Skip to content

The truly free, open-source, local-first, cross-platform CapCut replacement. Scriptable through a JSON-RPC API & MCP.

feat(speech): the sheet shows each model its own controls, and a sample voice

Jareerb7a1b912d899
diff --git a/src/crates/concat-speech/src/tts.rs b/src/crates/concat-speech/src/tts.rs
index a52db37f..07482430 100644
--- a/src/crates/concat-speech/src/tts.rs
+++ b/src/crates/concat-speech/src/tts.rs
@@ -103,12 +103,12 @@ const KNOWN_MODELS: &[KnownModel] = &[
KnownModel {
id: crate::chatterbox::BUNDLE_ID,
label: "Chatterbox Turbo (studio)",
- blurb: "The closest to a studio voice; reads in the selected clip's voice. English, slow, a gigabyte.",
+ blurb: "The closest to a studio voice; reads in Pocket's recordings or any sample of your own. English, slow, a gigabyte.",
approx_bytes: crate::chatterbox::BUNDLE_BYTES,
sha256: "",
},
KnownModel {
- id: "sherpa-onnx-pocket-tts-int8-2026-01-26",
+ id: POCKET_ID,
label: "Pocket TTS (cloning)",
blurb: "Reads in any voice, including one from a clip on the timeline. English.",
approx_bytes: 98_336_520,
@@ -187,17 +187,29 @@ pub const POCKET_VOICES: &[(i32, &str, Option<&str>)] = &[
(POCKET_CLONE, "pocket_clone", None),
];
-/// The voice that is the selected clip's own.
+/// The voice that is a recording of the caller's choosing - a clip on the
+/// timeline, a file in the bin - handed over as the request's reference.
pub const POCKET_CLONE: i32 = 1099;
-/// Chatterbox's voices: only the selected clip's own, since the model has
-/// no speakers of its own and ships no recording of anyone's.
-pub const CHATTERBOX_VOICES: &[(i32, &str)] = &[(CHATTERBOX_CLONE, "chatterbox_clone")];
+/// The Pocket bundle, which is also where the stock recordings live.
+pub const POCKET_ID: &str = "sherpa-onnx-pocket-tts-int8-2026-01-26";
+
+/// Chatterbox's voices. The model has no speakers of its own and ships no
+/// recording of anyone's under a licence this build may carry, so its
+/// stock voices are Pocket's two recordings, read from the Pocket bundle:
+/// offered only while that bundle is on disk, and named as Chatterbox's
+/// so a voice id still says which family reads it. The third is a
+/// recording of the caller's choosing, as Pocket's is.
+pub const CHATTERBOX_VOICES: &[(i32, &str, Option<&str>)] = &[
+ (2000, "chatterbox_bria", Some("test_wavs/bria.wav")),
+ (2001, "chatterbox_loona", Some("test_wavs/loona.wav")),
+ (CHATTERBOX_CLONE, "chatterbox_clone", None),
+];
-/// The selected clip's own voice, read by Chatterbox.
+/// A recording of the caller's choosing, read by Chatterbox.
pub const CHATTERBOX_CLONE: i32 = 2099;
-/// Whether a voice is the selected clip's own, in any family.
+/// Whether a voice is a recording of the caller's choosing, in any family.
pub fn is_clone(voice: i32) -> bool {
voice == POCKET_CLONE || voice == CHATTERBOX_CLONE
}
@@ -212,7 +224,7 @@ pub fn voice_family(voice: i32) -> Option<Family> {
Some(Family::Kokoro)
} else if POCKET_VOICES.iter().any(|(id, _, _)| *id == voice) {
Some(Family::Pocket)
- } else if CHATTERBOX_VOICES.iter().any(|(id, _)| *id == voice) {
+ } else if CHATTERBOX_VOICES.iter().any(|(id, _, _)| *id == voice) {
Some(Family::Chatterbox)
} else {
None
@@ -234,8 +246,8 @@ fn voice_name(voice: i32) -> Option<&'static str> {
.or_else(|| {
CHATTERBOX_VOICES
.iter()
- .find(|(id, _)| *id == voice)
- .map(|(_, name)| *name)
+ .find(|(id, _, _)| *id == voice)
+ .map(|(_, name, _)| *name)
})
}
@@ -362,8 +374,19 @@ pub struct SpeakRequest {
pub reference: Option<Reference>,
/// What to say.
pub text: String,
- /// Speaking rate; 1.0 is the voice's natural pace.
+ /// Speaking rate; 1.0 is the voice's natural pace. Kokoro and Pocket
+ /// honour it; Chatterbox reads at its own.
pub speed: f32,
+ /// The breath between sentences, as sherpa scales it: 0 runs them
+ /// together, 1 is a long pause, 0.2 is the engine's own. Absent is
+ /// the engine's own. Kokoro and Pocket.
+ #[serde(default)]
+ pub pauses: Option<f32>,
+ /// Pocket's flow-matching steps: more is a cleaner voice and a longer
+ /// wait, 5 is the engine's own. Absent is the engine's own; the other
+ /// families have no such dial and ignore it.
+ #[serde(default)]
+ pub steps: Option<i32>,
/// The project folder the WAV should land in.
pub project: String,
}
@@ -449,11 +472,18 @@ impl Speech {
name: (*name).to_owned(),
family: Family::Pocket,
}))
- .chain(CHATTERBOX_VOICES.iter().map(|(id, name)| VoiceInfo {
- id: *id,
- name: (*name).to_owned(),
- family: Family::Chatterbox,
- }))
+ // Chatterbox's stock voices are Pocket's recordings, and are
+ // only offered while the bundle holding them is here.
+ .chain(
+ CHATTERBOX_VOICES
+ .iter()
+ .filter(|(_, _, file)| file.is_none() || model_downloaded(dirs, POCKET_ID))
+ .map(|(id, name, _)| VoiceInfo {
+ id: *id,
+ name: (*name).to_owned(),
+ family: Family::Chatterbox,
+ }),
+ )
.collect(),
})
}
@@ -607,6 +637,14 @@ impl Speech {
} else {
1.0
};
+ let defaults = GenerationConfig::default();
+ let silence_scale = request
+ .pauses
+ .filter(|pauses| pauses.is_finite())
+ .map_or(defaults.silence_scale, |pauses| pauses.clamp(0.0, 1.0));
+ let num_steps = request
+ .steps
+ .map_or(defaults.num_steps, |steps| steps.clamp(1, 16));
let dir = model_dir(dirs, &request.model_id)?;
if !model_downloaded(dirs, &request.model_id) {
@@ -664,10 +702,18 @@ impl Speech {
// Chatterbox is its own loop: the recording in, the samples out,
// and the file written here. Sherpa's families go on below.
+ // A named voice's recording lives in the Pocket bundle, whichever
+ // model reads it; a chosen recording needs no bundle at all.
+ let recordings = if is_clone(request.voice) {
+ dir.clone()
+ } else {
+ recordings_dir(dirs)?
+ };
+
#[cfg(feature = "chatterbox")]
let tts = match tts {
Engine::Chatterbox(chatterbox) => {
- let samples = reference_samples(&dir, request)?;
+ let samples = reference_samples(&recordings, request)?;
let mut progress = progress;
let spoken = chatterbox.speak(text, &samples, &cancel, &mut |fraction| {
progress(fraction);
@@ -694,12 +740,15 @@ impl Speech {
Family::Kokoro => GenerationConfig {
sid: request.voice,
speed,
+ silence_scale,
..Default::default()
},
_ => {
- let samples = reference_samples(&dir, request)?;
+ let samples = reference_samples(&recordings, request)?;
GenerationConfig {
speed,
+ silence_scale,
+ num_steps,
reference_audio: Some(samples),
reference_sample_rate: REFERENCE_RATE as i32,
extra: Some(
@@ -772,14 +821,28 @@ impl Speech {
/// the rate its bundled recordings are in.
const REFERENCE_RATE: u32 = 24_000;
-/// The recording a Pocket request reads the voice from, as mono samples
-/// at [`REFERENCE_RATE`]: one of the bundle's own for a named voice, or
-/// the request's reference for the clone, cut to [`REFERENCE_SECONDS`]
-/// from where the voice is heard.
+/// Where the stock recordings are: the Pocket bundle. An error naming the
+/// fix when it is not on disk, since a Chatterbox voice borrowed from it
+/// can be asked for without Pocket ever having been downloaded.
+fn recordings_dir(dirs: &AppDirs) -> Result<PathBuf, String> {
+ if !model_downloaded(dirs, POCKET_ID) {
+ return Err(
+ "the stock voices are Pocket TTS's recordings - download Pocket TTS in Settings > \
+ Speech, or read in a sample voice of your own"
+ .to_owned(),
+ );
+ }
+ model_dir(dirs, POCKET_ID)
+}
+
+/// The recording a request reads the voice from, as mono samples at
+/// [`REFERENCE_RATE`]: one of the stock recordings in `dir` for a named
+/// voice, or the request's reference for a chosen one, cut to
+/// [`REFERENCE_SECONDS`] from where the voice is heard.
fn reference_samples(dir: &Path, request: &SpeakRequest) -> Result<Vec<f32>, String> {
let (path, start) = if is_clone(request.voice) {
let reference = request.reference.as_ref().ok_or_else(|| {
- "no clip to take the voice from - select a clip with a voice in it".to_owned()
+ "no recording to take the voice from - pick a sample with a voice in it".to_owned()
})?;
(
PathBuf::from(&reference.path),
@@ -792,6 +855,7 @@ fn reference_samples(dir: &Path, request: &SpeakRequest) -> Result<Vec<f32>, Str
} else {
let file = POCKET_VOICES
.iter()
+ .chain(CHATTERBOX_VOICES.iter())
.find(|(id, _, _)| *id == request.voice)
.and_then(|(_, _, file)| *file)
.ok_or_else(|| format!("voice {} has no recording", request.voice))?;
@@ -956,11 +1020,15 @@ mod tests {
}
for (id, name, _) in POCKET_VOICES {
assert_eq!(voice_name(*id), Some(*name));
- assert!(!CHATTERBOX_VOICES.iter().any(|(other, _)| other == id));
+ assert!(!CHATTERBOX_VOICES.iter().any(|(other, _, _)| other == id));
}
- for (id, name) in CHATTERBOX_VOICES {
+ for (id, name, _) in CHATTERBOX_VOICES {
assert_eq!(voice_name(*id), Some(*name));
+ assert_eq!(voice_family(*id), Some(Family::Chatterbox));
}
+ // The stock Chatterbox voices are Pocket's recordings by another name.
+ assert_eq!(CHATTERBOX_VOICES[0].2, POCKET_VOICES[0].2);
+ assert_eq!(CHATTERBOX_VOICES[1].2, POCKET_VOICES[1].2);
assert_eq!(voice_name(999), None);
for model in KNOWN_MODELS {
assert!(known(model.id).is_some());
@@ -977,10 +1045,12 @@ mod tests {
reference,
text: "hello".to_owned(),
speed: 1.0,
+ pauses: None,
+ steps: None,
project: String::new(),
};
let error = reference_samples(&dir, &request(POCKET_CLONE, None)).expect_err("no clip");
- assert!(error.contains("select a clip"), "{error}");
+ assert!(error.contains("pick a sample"), "{error}");
// A recording that is not there names itself.
let error = reference_samples(&dir, &request(1000, None)).expect_err("no file");
assert!(error.contains("bria.wav"), "{error}");
diff --git a/src/crates/concat/locales/de.json b/src/crates/concat/locales/de.json
index de764ec5..8b9a1ba0 100644
--- a/src/crates/concat/locales/de.json
+++ b/src/crates/concat/locales/de.json
@@ -203,7 +203,6 @@
"Fade in": "Einblenden",
"Fade out": "Ausblenden",
"Fades": "Blenden",
- "Faster": "Schneller",
"Favourites": "Favoriten",
"Feather": "Weiche Kante",
"female": "weiblich",
@@ -358,7 +357,6 @@
"Outline": "Kontur",
"Output": "Ausgabe",
"Output · {0}": "Ausgabe · {0}",
- "Pace": "Tempo",
"Paste": "Einfügen",
"Pause": "Pause",
"Person": "Person",
@@ -449,7 +447,6 @@
"Slide Left": "Nach links gleiten",
"Slide Right": "Nach rechts gleiten",
"Slide Up": "Nach oben gleiten",
- "Slower": "Langsamer",
"Small": "Klein",
"Smart brush": "Smart-Pinsel",
"Smart brush: {0}": "Smart-Pinsel: {0}",
diff --git a/src/crates/concat/locales/en.json b/src/crates/concat/locales/en.json
index 079a8cbd..29f341c6 100644
--- a/src/crates/concat/locales/en.json
+++ b/src/crates/concat/locales/en.json
@@ -77,6 +77,7 @@
"Analysis": "Analysis",
"Angle": "Angle",
"Animation": "Animation",
+ "Any voice, from a recording": "Any voice, from a recording",
"Appearance": "Appearance",
"Application": "Application",
"Apply": "Apply",
@@ -136,6 +137,7 @@
"Browse": "Browse",
"Brush": "Brush",
"Build": "Build",
+ "Built-in speakers · fast": "Built-in speakers · fast",
"Bulge / Pinch": "Bulge / Pinch",
"Bullet": "Bullet",
"Bypass {0}": "Bypass {0}",
@@ -153,6 +155,7 @@
"centred": "centred",
"Change pitch with speed": "Change pitch with speed",
"Change what this panel shows, or drag it onto another": "Change what this panel shows, or drag it onto another",
+ "Chatterbox has no voice of its own. Turn on Use sample voice below, or download Pocket TTS in Settings › Speech to borrow its two recordings.": "Chatterbox has no voice of its own. Turn on Use sample voice below, or download Pocket TTS in Settings › Speech to borrow its two recordings.",
"Chill": "Chill",
"Chinese": "Chinese",
"Chipmunk": "Chipmunk",
@@ -310,6 +313,7 @@
"Export…": "Export…",
"Exposure": "Exposure",
"Exposure is in stops; the rest is a signed percent of the picture as shot, and zero everywhere is the picture as shot. Double-click a label to put it back; the diamond keys the knob at the playhead.": "Exposure is in stops; the rest is a signed percent of the picture as shot, and zero everywhere is the picture as shot. Double-click a label to put it back; the diamond keys the knob at the playhead.",
+ "Expression": "Expression",
"Faces brightened and evened: skin lifted, reds calmed, the balance neutral, the rest of the frame left where it was.": "Faces brightened and evened: skin lifted, reds calmed, the balance neutral, the rest of the frame left where it was.",
"Fade": "Fade",
"Fade in": "Fade in",
@@ -318,7 +322,6 @@
"Fades": "Fades",
"Fair": "Fair",
"Falling Dust": "Falling Dust",
- "Faster": "Faster",
"Favourites": "Favourites",
"Feather": "Feather",
"Featured": "Featured",
@@ -334,6 +337,7 @@
"Filters": "Filters",
"Finalising file": "Finalising file",
"Finding the subject… {0}%": "Finding the subject… {0}%",
+ "Fine": "Fine",
"Fine specks of dust flicker over the picture, frame to frame.": "Fine specks of dust flicker over the picture, frame to frame.",
"Fisheye": "Fisheye",
"Fit the frame to the pane": "Fit the frame to the pane",
@@ -417,6 +421,7 @@
"Images": "Images",
"Impact Ring": "Impact Ring",
"Import": "Import",
+ "Import a clip with a voice in it first.": "Import a clip with a voice in it first.",
"Import a LUT (.cube)": "Import a LUT (.cube)",
"Import LUT": "Import LUT",
"Import media": "Import media",
@@ -483,6 +488,8 @@
"Location": "Location",
"Lock lane": "Lock lane",
"Lock track": "Lock track",
+ "long": "long",
+ "longer": "longer",
"Loop": "Loop",
"Low contrast, pink and lavender in the light, and a soft bloom over it all. The dreamy one.": "Low contrast, pink and lavender in the light, and a soft bloom over it all. The dreamy one.",
"Low light opened up: shadows lifted with the noise softened before it rises, whites held, the blue of the night kept.": "Low light opened up: shadows lifted with the noise softened before it rises, whites held, the blue of the night kept.",
@@ -540,6 +547,7 @@
"My templates": "My templates",
"Name": "Name",
"Natural": "Natural",
+ "natural": "natural",
"Neon": "Neon",
"Neon Edges": "Neon Edges",
"Neon Flicker": "Neon Flicker",
@@ -597,15 +605,17 @@
"Outline": "Outline",
"Output": "Output",
"Output · {0}": "Output · {0}",
- "Pace": "Pace",
"Pan tool (H)": "Pan tool (H)",
"Party & Club": "Party & Club",
"Paste": "Paste",
"Pause": "Pause",
+ "Pauses": "Pauses",
"Peach": "Peach",
"Period": "Period",
"Periodic horizontal slice displacement with chromatic aberration.": "Periodic horizontal slice displacement with chromatic aberration.",
"Person": "Person",
+ "Pick a recording": "Pick a recording",
+ "Pick a sample with a voice in it": "Pick a sample with a voice in it",
"Pick an effect or filter from the library": "Pick an effect or filter from the library",
"Pick one from the Transitions tab to dissolve into this clip from the cut before it.": "Pick one from the Transitions tab to dissolve into this clip from the cut before it.",
"Pink Fuzz": "Pink Fuzz",
@@ -644,6 +654,7 @@
"Push": "Push",
"Quality": "Quality",
"Quarter": "Quarter",
+ "Quick": "Quick",
"Radial chromatic dispersion refracting RGB channels.": "Radial chromatic dispersion refracting RGB channels.",
"Radial ripple expansion pulsing with beat intensity.": "Radial ripple expansion pulsing with beat intensity.",
"Radio": "Radio",
@@ -655,6 +666,8 @@
"Rapid inward/outward camera rush with radial zoom blur.": "Rapid inward/outward camera rush with radial zoom blur.",
"Rate": "Rate",
"Rate re-times the clip in place, so the timeline length changes with it. Trim afterwards, not before. A curve varies the rate over the clip; the same footage plays, faster here and slower there.": "Rate re-times the clip in place, so the timeline length changes with it. Trim afterwards, not before. A curve varies the rate over the clip; the same footage plays, faster here and slower there.",
+ "Reads in any voice from a few seconds of a recording: one it ships with, or a sample from the bin.": "Reads in any voice from a few seconds of a recording: one it ships with, or a sample from the bin.",
+ "Reads in the voice heard in a recording from the bin. The first ten seconds from the point chosen below are what the model listens to.": "Reads in the voice heard in a recording from the bin. The first ten seconds from the point chosen below are what the model listens to.",
"Reads in the voice heard in the clip selected on the timeline": "Reads in the voice heard in the clip selected on the timeline",
"Reattach audio": "Reattach audio",
"Recent": "Recent",
@@ -702,6 +715,7 @@
"Row height": "Row height",
"Runs locally": "Runs locally",
"Sage": "Sage",
+ "Sample": "Sample",
"Saturated and bright, the water pushed teal.": "Saturated and bright, the water pushed teal.",
"Saturation": "Saturation",
"Save": "Save",
@@ -722,7 +736,6 @@
"See all": "See all",
"Segments": "Segments",
"Select a clip on the timeline first": "Select a clip on the timeline first",
- "Select a clip with a voice in it to clone": "Select a clip with a voice in it to clone",
"Select a video or image clip on the timeline first": "Select a video or image clip on the timeline first",
"Select a video, image or text clip on the timeline first": "Select a video, image or text clip on the timeline first",
"Select folder containing media files": "Select folder containing media files",
@@ -765,7 +778,6 @@
"Slide Left": "Slide Left",
"Slide Right": "Slide Right",
"Slide Up": "Slide Up",
- "Slower": "Slower",
"Small": "Small",
"Smallest": "Smallest",
"Smart brush": "Smart brush",
@@ -790,6 +802,7 @@
"Space": "Space",
"Spanish": "Spanish",
"Sparkle": "Sparkle",
+ "Speakers built into the model, in English and Chinese. Reads at the speed and with the pauses set below.": "Speakers built into the model, in English and Chinese. Reads at the speed and with the pauses set below.",
"Speech": "Speech",
"Speed": "Speed",
"Spin": "Spin",
@@ -804,6 +817,7 @@
"Strength": "Strength",
"Strobe Flash": "Strobe Flash",
"Stroke": "Stroke",
+ "Studio voice · slow": "Studio voice · slow",
"Style": "Style",
"Stylised": "Stylised",
"Stylize": "Stylize",
@@ -838,6 +852,7 @@
"The centre of the picture bulges outward, or pinches inward below zero.": "The centre of the picture bulges outward, or pinches inward below zero.",
"The chain runs top to bottom — reorder it with the chevrons. The eye bypasses one link without losing the values you dialled into it.": "The chain runs top to bottom — reorder it with the chevrons. The eye bypasses one link without losing the values you dialled into it.",
"The clip is no longer on the timeline": "The clip is no longer on the timeline",
+ "The closest to a studio voice. Reads at its own pace, in one of Pocket's two recordings or in any sample from the bin, and reads [laugh], [chuckle] and [sigh] as what they say.": "The closest to a studio voice. Reads at its own pace, in one of Pocket's two recordings or in any sample from the bin, and reads [laugh], [chuckle] and [sigh] as what they say.",
"The colour channels pull apart through the cut and snap back together.": "The colour channels pull apart through the cut and snap back together.",
"The colour channels pulled apart into a bold stylised fringe.": "The colour channels pulled apart into a bold stylised fringe.",
"The cut twists both pictures around its centre and untwists into the next shot.": "The cut twists both pictures around its centre and untwists into the next shot.",
@@ -873,6 +888,7 @@
"These settings apply to the whole edit and cannot be changed later.": "These settings apply to the whole edit and cannot be changed later.",
"This clip has no file to transcribe": "This clip has no file to transcribe",
"This file has no picture to enhance": "This file has no picture to enhance",
+ "This model reads only with its own speakers": "This model reads only with its own speakers",
"Threshold": "Threshold",
"Tiles": "Tiles",
"Tilt Shift": "Tilt Shift",
@@ -917,6 +933,7 @@
"Unpacking…": "Unpacking…",
"Untitled project": "Untitled project",
"Up or down by semitones, the length untouched.": "Up or down by semitones, the length untouched.",
+ "Use sample voice": "Use sample voice",
"Vaporwave": "Vaporwave",
"Vast stadium echoes and reverberations across a massive arena.": "Vast stadium echoes and reverberations across a massive arena.",
"VBR": "VBR",
@@ -939,6 +956,7 @@
"Vivid Landscape": "Vivid Landscape",
"Voice": "Voice",
"Voice added to the timeline": "Voice added to the timeline",
+ "Voice from": "Voice from",
"Voices": "Voices",
"Volume": "Volume",
"Vortex Swirl": "Vortex Swirl",
@@ -984,6 +1002,7 @@
"{0} media file(s) not found": "{0} media file(s) not found",
"{0} minutes ago": "{0} minutes ago",
"{0} months ago": "{0} months ago",
+ "{0} s in": "{0} s in",
"{0} strokes": "{0} strokes",
"{0}m {1}s left": "{0}m {1}s left",
"{0}s left": "{0}s left"
diff --git a/src/crates/concat/locales/es.json b/src/crates/concat/locales/es.json
index 4d05bb0f..2e9387ba 100644
--- a/src/crates/concat/locales/es.json
+++ b/src/crates/concat/locales/es.json
@@ -203,7 +203,6 @@
"Fade in": "Fundido de entrada",
"Fade out": "Fundido de salida",
"Fades": "Fundidos",
- "Faster": "Más rápido",
"Favourites": "Favoritos",
"Feather": "Difuminado",
"female": "femenina",
@@ -358,7 +357,6 @@
"Outline": "Contorno",
"Output": "Salida",
"Output · {0}": "Salida · {0}",
- "Pace": "Ritmo",
"Paste": "Pegar",
"Pause": "Pausa",
"Person": "Persona",
@@ -449,7 +447,6 @@
"Slide Left": "Deslizar a la izquierda",
"Slide Right": "Deslizar a la derecha",
"Slide Up": "Deslizar arriba",
- "Slower": "Más lento",
"Small": "Pequeño",
"Smart brush": "Pincel inteligente",
"Smart brush: {0}": "Pincel inteligente: {0}",
diff --git a/src/crates/concat/locales/fa.json b/src/crates/concat/locales/fa.json
index 6befb7f4..47d57e55 100644
--- a/src/crates/concat/locales/fa.json
+++ b/src/crates/concat/locales/fa.json
@@ -234,7 +234,6 @@
"Fades": "محوها",
"Fair": "متعادل",
"Falling Dust": "گردوغبار در حال سقوط",
- "Faster": "سریع‌تر",
"Favourites": "علاقه‌مندی‌ها",
"Feather": "نرمی لبه",
"female": "زن",
@@ -443,7 +442,6 @@
"Outline": "حاشیه",
"Output": "خروجی",
"Output · {0}": "خروجی · {0}",
- "Pace": "سرعت",
"Pan tool (H)": "ابزار Pan (H)",
"Turn on Preview Axis (S)": "روشن کردن محور پیش‌نمایش (S)",
"Turn off Preview Axis (S)": "خاموش کردن محور پیش‌نمایش (S)",
@@ -572,7 +570,6 @@
"Slide Left": "اسلاید به چپ",
"Slide Right": "اسلاید به راست",
"Slide Up": "اسلاید به بالا",
- "Slower": "کندتر",
"Small": "کوچک",
"Smallest": "کوچک‌ترین",
"Smart brush": "قلم‌موی هوشمند",
diff --git a/src/crates/concat/locales/fr.json b/src/crates/concat/locales/fr.json
index 14f04826..2574925e 100644
--- a/src/crates/concat/locales/fr.json
+++ b/src/crates/concat/locales/fr.json
@@ -203,7 +203,6 @@
"Fade in": "Fondu d'entrée",
"Fade out": "Fondu de sortie",
"Fades": "Fondus",
- "Faster": "Plus rapide",
"Favourites": "Favoris",
"Feather": "Contour progressif",
"female": "féminine",
@@ -358,7 +357,6 @@
"Outline": "Contour",
"Output": "Sortie",
"Output · {0}": "Sortie · {0}",
- "Pace": "Rythme",
"Paste": "Coller",
"Pause": "Pause",
"Person": "Personne",
@@ -449,7 +447,6 @@
"Slide Left": "Glisser vers la gauche",
"Slide Right": "Glisser vers la droite",
"Slide Up": "Glisser vers le haut",
- "Slower": "Plus lent",
"Small": "Petit",
"Smart brush": "Pinceau intelligent",
"Smart brush: {0}": "Pinceau intelligent : {0}",
diff --git a/src/crates/concat/locales/hr.json b/src/crates/concat/locales/hr.json
index b1a06299..331bf53b 100644
--- a/src/crates/concat/locales/hr.json
+++ b/src/crates/concat/locales/hr.json
@@ -250,7 +250,6 @@
"Fades": "Izblijedjenja",
"Fair": "Svijetlo",
"Falling Dust": "Padajuća prašina",
- "Faster": "Brže",
"Favourites": "Favoriti",
"Feather": "Omekšanje",
"female": "ženski",
@@ -445,7 +444,6 @@
"Outline": "Kontura",
"Output": "Izlaz",
"Output · {0}": "Izlaz · {0}",
- "Pace": "Tempo",
"Paste": "Zalijepi",
"Pause": "Pauza",
"Peach": "Breskva",
@@ -558,7 +556,6 @@
"Slide Left": "Klizanje lijevo",
"Slide Right": "Klizanje desno",
"Slide Up": "Klizanje gore",
- "Slower": "Sporije",
"Small": "Malo",
"Smart brush": "Pametna četka",
"Smart brush: {0}": "Pametna četka: {0}",
diff --git a/src/crates/concat/locales/it.json b/src/crates/concat/locales/it.json
index 020ac784..60af0ae6 100644
--- a/src/crates/concat/locales/it.json
+++ b/src/crates/concat/locales/it.json
@@ -203,7 +203,6 @@
"Fade in": "Dissolvenza in entrata",
"Fade out": "Dissolvenza in uscita",
"Fades": "Dissolvenze",
- "Faster": "Più veloce",
"Favourites": "Preferiti",
"Feather": "Sfumatura",
"female": "femminile",
@@ -358,7 +357,6 @@
"Outline": "Contorno",
"Output": "Uscita",
"Output · {0}": "Uscita · {0}",
- "Pace": "Ritmo",
"Paste": "Incolla",
"Pause": "Pausa",
"Person": "Persona",
@@ -449,7 +447,6 @@
"Slide Left": "Scorri a sinistra",
"Slide Right": "Scorri a destra",
"Slide Up": "Scorri in alto",
- "Slower": "Più lento",
"Small": "Piccolo",
"Smart brush": "Pennello intelligente",
"Smart brush: {0}": "Pennello intelligente: {0}",
diff --git a/src/crates/concat/locales/ja.json b/src/crates/concat/locales/ja.json
index 5a2cea99..590b7008 100644
--- a/src/crates/concat/locales/ja.json
+++ b/src/crates/concat/locales/ja.json
@@ -203,7 +203,6 @@
"Fade in": "フェードイン",
"Fade out": "フェードアウト",
"Fades": "フェード",
- "Faster": "速く",
"Favourites": "お気に入り",
"Feather": "ぼかし",
"female": "女性",
@@ -358,7 +357,6 @@
"Outline": "縁取り",
"Output": "出力",
"Output · {0}": "出力 · {0}",
- "Pace": "速さ",
"Paste": "ペースト",
"Pause": "一時停止",
"Person": "人物",
@@ -449,7 +447,6 @@
"Slide Left": "左へスライド",
"Slide Right": "右へスライド",
"Slide Up": "上へスライド",
- "Slower": "遅く",
"Small": "小",
"Smart brush": "スマートブラシ",
"Smart brush: {0}": "スマートブラシ: {0}",
diff --git a/src/crates/concat/locales/ko.json b/src/crates/concat/locales/ko.json
index 344b4317..5e4d720b 100644
--- a/src/crates/concat/locales/ko.json
+++ b/src/crates/concat/locales/ko.json
@@ -203,7 +203,6 @@
"Fade in": "페이드 인",
"Fade out": "페이드 아웃",
"Fades": "페이드",
- "Faster": "빠르게",
"Favourites": "즐겨찾기",
"Feather": "페더",
"female": "여성",
@@ -358,7 +357,6 @@
"Outline": "외곽선",
"Output": "출력",
"Output · {0}": "출력 · {0}",
- "Pace": "속도",
"Paste": "붙여넣기",
"Pause": "일시 정지",
"Person": "인물",
@@ -449,7 +447,6 @@
"Slide Left": "왼쪽으로 슬라이드",
"Slide Right": "오른쪽으로 슬라이드",
"Slide Up": "위로 슬라이드",
- "Slower": "느리게",
"Small": "작게",
"Smart brush": "스마트 브러시",
"Smart brush: {0}": "스마트 브러시: {0}",
diff --git a/src/crates/concat/locales/pt-BR.json b/src/crates/concat/locales/pt-BR.json
index fab8b096..4b99464a 100644
--- a/src/crates/concat/locales/pt-BR.json
+++ b/src/crates/concat/locales/pt-BR.json
@@ -203,7 +203,6 @@
"Fade in": "Fade de entrada",
"Fade out": "Fade de saída",
"Fades": "Fades",
- "Faster": "Mais rápido",
"Favourites": "Favoritos",
"Feather": "Suavização",
"female": "feminina",
@@ -358,7 +357,6 @@
"Outline": "Contorno",
"Output": "Saída",
"Output · {0}": "Saída · {0}",
- "Pace": "Ritmo",
"Paste": "Colar",
"Pause": "Pausar",
"Person": "Pessoa",
@@ -449,7 +447,6 @@
"Slide Left": "Deslizar para a esquerda",
"Slide Right": "Deslizar para a direita",
"Slide Up": "Deslizar para cima",
- "Slower": "Mais lento",
"Small": "Pequeno",
"Smart brush": "Pincel inteligente",
"Smart brush: {0}": "Pincel inteligente: {0}",
diff --git a/src/crates/concat/locales/ru.json b/src/crates/concat/locales/ru.json
index 49eb4df0..c35de0ff 100644
--- a/src/crates/concat/locales/ru.json
+++ b/src/crates/concat/locales/ru.json
@@ -203,7 +203,6 @@
"Fade in": "Нарастание",
"Fade out": "Затухание",
"Fades": "Плавные переходы",
- "Faster": "Быстрее",
"Favourites": "Избранное",
"Feather": "Растушёвка",
"female": "женский",
@@ -358,7 +357,6 @@
"Outline": "Обводка",
"Output": "Вывод",
"Output · {0}": "Вывод · {0}",
- "Pace": "Темп",
"Paste": "Вставить",
"Pause": "Пауза",
"Person": "Человек",
@@ -449,7 +447,6 @@
"Slide Left": "Сдвиг влево",
"Slide Right": "Сдвиг вправо",
"Slide Up": "Сдвиг вверх",
- "Slower": "Медленнее",
"Small": "Малый",
"Smart brush": "Умная кисть",
"Smart brush: {0}": "Умная кисть: {0}",
diff --git a/src/crates/concat/locales/tr.json b/src/crates/concat/locales/tr.json
index f831e7a1..5bb39896 100644
--- a/src/crates/concat/locales/tr.json
+++ b/src/crates/concat/locales/tr.json
@@ -203,7 +203,6 @@
"Fade in": "Açılma",
"Fade out": "Kararma",
"Fades": "Geçişler",
- "Faster": "Daha hızlı",
"Favourites": "Favoriler",
"Feather": "Yumuşatma",
"female": "kadın",
@@ -358,7 +357,6 @@
"Outline": "Dış çizgi",
"Output": "Çıktı",
"Output · {0}": "Çıkış · {0}",
- "Pace": "Tempo",
"Paste": "Yapıştır",
"Pause": "Duraklat",
"Person": "Kişi",
@@ -449,7 +447,6 @@
"Slide Left": "Sola kay",
"Slide Right": "Sağa kay",
"Slide Up": "Yukarı kay",
- "Slower": "Daha yavaş",
"Small": "Küçük",
"Smart brush": "Akıllı fırça",
"Smart brush: {0}": "Akıllı fırça: {0}",
diff --git a/src/crates/concat/locales/zh-Hans.json b/src/crates/concat/locales/zh-Hans.json
index 1bbf23ed..4074df7a 100644
--- a/src/crates/concat/locales/zh-Hans.json
+++ b/src/crates/concat/locales/zh-Hans.json
@@ -203,7 +203,6 @@
"Fade in": "淡入",
"Fade out": "淡出",
"Fades": "淡入淡出",
- "Faster": "更快",
"Favourites": "收藏",
"Feather": "羽化",
"female": "女声",
@@ -358,7 +357,6 @@
"Outline": "描边",
"Output": "输出",
"Output · {0}": "输出 · {0}",
- "Pace": "语速",
"Paste": "粘贴",
"Pause": "暂停",
"Person": "人物",
@@ -449,7 +447,6 @@
"Slide Left": "左滑",
"Slide Right": "右滑",
"Slide Up": "上滑",
- "Slower": "更慢",
"Small": "小",
"Smart brush": "智能画笔",
"Smart brush: {0}": "智能画笔:{0}",
diff --git a/src/crates/concat/locales/zh-TW.json b/src/crates/concat/locales/zh-TW.json
index 174231d7..634ff129 100644
--- a/src/crates/concat/locales/zh-TW.json
+++ b/src/crates/concat/locales/zh-TW.json
@@ -314,7 +314,6 @@
"Fades": "淡入淡出",
"Fair": "透亮",
"Falling Dust": "飄塵",
- "Faster": "較快",
"Favourites": "收藏",
"Feather": "羽化",
"Featured": "精選",
@@ -589,7 +588,6 @@
"Outline": "外框",
"Output": "輸出",
"Output · {0}": "輸出 · {0}",
- "Pace": "語速",
"Pan tool (H)": "平移工具 (H)",
"Turn on Preview Axis (S)": "開啟預覽軸 (S)",
"Turn off Preview Axis (S)": "關閉預覽軸 (S)",
@@ -714,7 +712,6 @@
"Search {0}s": "搜尋{0}",
"Segments": "分段數",
"Select a clip on the timeline first": "請先在時間軸上選取片段",
- "Select a clip with a voice in it to clone": "請選取一段帶有人聲的片段,以複製其聲音",
"Select a video or image clip on the timeline first": "請先選取時間軸上的影片或圖片片段",
"Select a video, image or text clip on the timeline first": "請先在時間軸上選取影片、圖片或文字片段",
"Select folder containing media files": "選擇包含媒體檔案的資料夾",
@@ -756,7 +753,6 @@
"Slide Left": "向左滑動",
"Slide Right": "向右滑動",
"Slide Up": "向上滑動",
- "Slower": "較慢",
"Small": "小",
"Smallest": "最小",
"Smart brush": "智慧筆刷",
diff --git a/src/crates/concat/src/lib.rs b/src/crates/concat/src/lib.rs
index f639fa2a..93e9ac86 100644
--- a/src/crates/concat/src/lib.rs
+++ b/src/crates/concat/src/lib.rs
@@ -199,8 +199,12 @@ pub fn run() -> Result<(), slint::PlatformError> {
editor.set_menu_items(ModelRc::from(models.menu.clone()));
app.set_caption_models(ModelRc::from(models.caption_models.clone()));
app.set_speech_models(ModelRc::from(models.speech_models.clone()));
+ app.set_speech_model_details(ModelRc::from(models.speech_model_details.clone()));
app.set_speech_voices(ModelRc::from(models.speakers.clone()));
app.set_speech_voice_details(ModelRc::from(models.speaker_details.clone()));
+ app.set_speech_samples(ModelRc::from(models.speech_samples.clone()));
+ app.set_speech_sample_waves(ModelRc::from(models.speech_sample_waves.clone()));
+ app.set_speech_sample_details(ModelRc::from(models.speech_sample_details.clone()));
app.set_app_menu_items(ModelRc::from(models.bar.clone()));
app.set_transcribers(ModelRc::from(models.transcribers.clone()));
app.set_voices(ModelRc::from(models.voices.clone()));
@@ -1272,9 +1276,29 @@ pub fn run() -> Result<(), slint::PlatformError> {
app.on_speech_model_changed(on_window!(|state, index: i32| {
state.handle(Msg::Speech(SpeechMsg::ModelChanged(index)));
}));
- app.on_speech_pace_changed(on_window!(|state, index: i32| {
- state.handle(Msg::Speech(SpeechMsg::PaceChanged(index)));
+ app.on_speech_speed_changed(on_window!(|state, value: f32| {
+ state.handle(Msg::Speech(SpeechMsg::SpeedChanged(value)));
}));
+ app.on_speech_quality_changed(on_window!(|state, index: i32| {
+ state.handle(Msg::Speech(SpeechMsg::QualityChanged(index)));
+ }));
+ app.on_speech_pauses_changed(on_window!(|state, value: f32| {
+ state.handle(Msg::Speech(SpeechMsg::PausesChanged(value)));
+ }));
+ app.on_speech_reference_start_changed(on_window!(|state, seconds: f32| {
+ state.handle(Msg::Speech(SpeechMsg::ReferenceStartChanged(seconds)));
+ }));
+ app.on_speech_use_sample_changed(on_window!(|state, on: bool| {
+ state.handle(Msg::Speech(SpeechMsg::UseSampleChanged(on)));
+ }));
+ app.on_speech_sample_changed(on_window!(|state, index: i32| {
+ state.handle(Msg::Speech(SpeechMsg::SampleChanged(index)));
+ }));
+ // A string cut, and nothing of the window's: the sheet keeps the text,
+ // Rust only knows where a character starts.
+ app.on_speech_insert_tag(|text: SharedString, tag: SharedString, at: i32| {
+ panes::speech::insert_tag(text.as_str(), tag.as_str(), at.max(0) as usize).into()
+ });
app.on_speech_begin(on_window!(|state| {
state.handle(Msg::Speech(SpeechMsg::Begin));
}));
diff --git a/src/crates/concat/src/panes/speech.rs b/src/crates/concat/src/panes/speech.rs
index dd91f2ed..51f6ce56 100644
--- a/src/crates/concat/src/panes/speech.rs
+++ b/src/crates/concat/src/panes/speech.rs
@@ -8,15 +8,29 @@
//! chosen last time. The synthesiser runs on a worker and reports as
//! [`SpeechMsg::Progress`], ending as [`SpeechMsg::Finished`] with the
//! written file, which lands in the bin and on the timeline.
+//!
+//! The model comes first and decides the rest: which voices are offered,
+//! and which dials. Kokoro reads with built-in speakers at a speed and
+//! with a pause of your choosing; Pocket reads in a recording, at a speed,
+//! with a pause, and at a quality that is its flow steps; Chatterbox reads
+//! at its own pace and knows the expression tags. Either of the two that
+//! read in a recording can read in a sample voice instead of a named one:
+//! a recording from the bin, chosen by its waveform, from a point in it of
+//! the sheet's choosing. The sheet shows each model its own controls rather
+//! than the few every model shares, which is why the pane carries every
+//! dial and the family says which are read.
use std::sync::Arc;
use concat_host::media::{self, MediaSummary};
use concat_project::Command;
-use concat_project::model::{ClipKind, MediaOrigin};
-use concat_speech::tts::{CHATTERBOX_CLONE, Family, Reference, VoiceInfo, family_of, is_clone};
+use concat_project::model::{ClipKind, MediaItem, MediaOrigin};
+use concat_speech::tts::{
+ CHATTERBOX_CLONE, Family, POCKET_CLONE, Reference, VoiceInfo, family_of, is_clone,
+};
use slint::SharedString;
+use crate::format::wave_path;
use crate::host::{on_ui, spawn};
use crate::i18n::{t, tf};
use crate::panes::Msg;
@@ -29,8 +43,16 @@ use crate::ui::SpeechSheetData;
const DEFAULT_VOICE: i32 = 3;
/// The default Pocket voice: Bria, the bundle's first recording.
const DEFAULT_POCKET_VOICE: i32 = 1000;
-/// The paces the speech sheet offers, as the voice's rate multiplier.
-const PACES: [f32; 3] = [0.85, 1.0, 1.15];
+/// The default Chatterbox voice: the same recording, read by Chatterbox.
+const DEFAULT_CHATTERBOX_VOICE: i32 = 2000;
+/// The speeds a voice can be asked for, as the rate multiplier the sheet's
+/// slider runs over; 1 is the voice's own.
+const SPEEDS: (f32, f32) = (0.5, 2.0);
+/// Pocket's flow steps behind the sheet's three qualities: quick, the
+/// engine's own, and fine.
+const STEPS: [i32; 3] = [3, 5, 8];
+/// The breath between sentences the sheet opens on: sherpa's own.
+const DEFAULT_PAUSES: f32 = 0.2;
/// Everything that can happen to the speech sheet.
#[derive(Clone, Debug)]
@@ -41,7 +63,18 @@ pub enum SpeechMsg {
TextEdited(String),
VoiceChanged(i32),
ModelChanged(i32),
- PaceChanged(i32),
+ /// The rate, `0.5..=2`.
+ SpeedChanged(f32),
+ /// Pocket's quality: a row of [`STEPS`].
+ QualityChanged(i32),
+ /// The breath between sentences, `0..=1`.
+ PausesChanged(f32),
+ /// Seconds into the sample the voice is taken from.
+ ReferenceStartChanged(f32),
+ /// Read in a sample from the bin rather than a named voice.
+ UseSampleChanged(bool),
+ /// Which sample: a row of [`SpeechPane::samples`].
+ SampleChanged(i32),
/// Read the script.
Begin,
Cancel,
@@ -64,10 +97,24 @@ pub struct SpeechPane {
pub voice: usize,
/// Row in the installed voice model list.
pub model: usize,
- /// The family of the chosen model, which is which voices are offered.
+ /// The family of the chosen model, which is which voices are offered
+ /// and which of the dials below are read.
pub family: Family,
- /// 0 slower, 1 natural, 2 faster.
- pub pace: usize,
+ /// The rate, `0.5..=2`; Kokoro and Pocket.
+ pub speed: f32,
+ /// Row of [`STEPS`]; Pocket.
+ pub quality: usize,
+ /// The breath between sentences, `0..=1`; Kokoro and Pocket.
+ pub pauses: f32,
+ /// Reading in a sample from the bin rather than a named voice; Pocket
+ /// and Chatterbox, the two that read in a recording.
+ pub use_sample: bool,
+ /// Row of [`SpeechPane::samples`]: which recording.
+ pub sample: usize,
+ /// Seconds into the sample the voice is taken from: the first ten
+ /// seconds from there are what the model listens to, and the start of
+ /// a recording is not always where the speaking is.
+ pub reference_start: f64,
pub running: bool,
pub progress: f32,
pub message: String,
@@ -103,13 +150,21 @@ impl SpeechPane {
voice: 0,
model,
family,
- pace: 1,
+ speed: 1.0,
+ quality: 1,
+ pauses: DEFAULT_PAUSES,
+ use_sample: false,
+ sample: 0,
+ reference_start: 0.0,
speakers: std::mem::take(&mut self.speakers),
..SpeechPane::default()
};
- // The voice chosen last time, if the model reads with it.
+ // The voice chosen last time, if the model reads with it;
+ // and the selected clip's recording as the sample to start
+ // from, since that is the voice most likely wanted.
let wanted = studio.prefs.tts_voice;
self.voice = self.voice_row(wanted);
+ self.sample = self.default_sample(studio);
}
SpeechMsg::Close => self.open = false,
SpeechMsg::TextEdited(text) => self.text = text,
@@ -125,7 +180,35 @@ impl SpeechPane {
.unwrap_or(Family::Kokoro);
self.voice = self.voice_row(chosen);
}
- SpeechMsg::PaceChanged(index) => self.pace = (index.max(0) as usize).min(2),
+ SpeechMsg::SpeedChanged(speed) => {
+ self.speed = if speed.is_finite() {
+ speed.clamp(SPEEDS.0, SPEEDS.1)
+ } else {
+ 1.0
+ }
+ }
+ SpeechMsg::QualityChanged(index) => {
+ self.quality = (index.max(0) as usize).min(STEPS.len() - 1)
+ }
+ SpeechMsg::PausesChanged(pauses) => {
+ self.pauses = if pauses.is_finite() {
+ pauses.clamp(0.0, 1.0)
+ } else {
+ DEFAULT_PAUSES
+ }
+ }
+ SpeechMsg::ReferenceStartChanged(seconds) => {
+ self.reference_start = if seconds.is_finite() {
+ f64::from(seconds.max(0.0))
+ } else {
+ 0.0
+ }
+ }
+ SpeechMsg::UseSampleChanged(on) => self.use_sample = on,
+ SpeechMsg::SampleChanged(index) => {
+ self.sample = index.max(0) as usize;
+ self.reference_start = 0.0;
+ }
SpeechMsg::Begin => self.run(studio),
SpeechMsg::Cancel => {
studio.host.speech.cancel();
@@ -182,34 +265,41 @@ impl SpeechPane {
self.message = t("Download a voice model in Settings › Speech first");
return;
};
- let Some(voice) = self.offered().get(self.voice).map(|speaker| speaker.id) else {
- self.message = t("No voice to read with");
- return;
- };
- // The clone reads in the voice of the selected clip: its file,
- // from where the clip starts in it.
- let reference = if is_clone(voice) {
- let recording = studio
- .sole_selection()
- .and_then(|id| studio.clip(&id))
- .filter(|clip| clip.kind == ClipKind::Video || clip.kind == ClipKind::Audio)
- .and_then(|clip| {
- studio
- .project()
- .media_by_id(&clip.media_id)
- .filter(|media| media.has_audio)
- .map(|media| Reference {
- path: media.path.clone(),
- start: clip.source_start,
- })
- });
- let Some(recording) = recording else {
- self.message = t("Select a clip with a voice in it to clone");
+ // A sample voice is a recording of the sheet's choosing, handed over
+ // as the request's reference under the family's own id for one.
+ // Otherwise the voice is one the model names - and Chatterbox may
+ // name none, since its stock voices are Pocket's recordings and
+ // are only there while Pocket is.
+ let (voice, reference) = if self.use_sample {
+ let Some(voice) = self.sample_voice() else {
+ self.message = t("This model reads only with its own speakers");
+ return;
+ };
+ let Some(sample) = self.sample_media(studio).cloned() else {
+ self.message = t("Pick a sample with a voice in it");
return;
};
- Some(recording)
+ let start = self.reference_offset(sample.duration.unwrap_or(0.0));
+ (
+ voice,
+ Some(Reference {
+ path: sample.path,
+ start,
+ }),
+ )
} else {
- None
+ let Some(voice) = self.offered().get(self.voice).map(|speaker| speaker.id) else {
+ self.message = if self.family == Family::Chatterbox {
+ t(
+ "Chatterbox has no voice of its own: turn on Use sample voice, or \
+ download Pocket TTS for its two recordings",
+ )
+ } else {
+ t("No voice to read with")
+ };
+ return;
+ };
+ (voice, None)
};
let Some(project) = studio
.session
@@ -218,9 +308,12 @@ impl SpeechPane {
else {
return;
};
- // Remembered: the voice chosen is the voice wanted next time.
- studio.prefs.tts_voice = Some(voice);
- studio.prefs.save(&studio.host.dirs);
+ // Remembered: the voice chosen is the voice wanted next time. A
+ // sample is a recording in one project, and is not.
+ if !self.use_sample {
+ studio.prefs.tts_voice = Some(voice);
+ studio.prefs.save(&studio.host.dirs);
+ }
self.landing = self
.clip
.as_ref()
@@ -232,7 +325,10 @@ impl SpeechPane {
voice,
reference,
text,
- speed: PACES[self.pace.min(2)],
+ speed: self.rate(),
+ pauses: self.reads_at_a_pace().then_some(self.pauses),
+ steps: (self.family == Family::Pocket)
+ .then(|| STEPS[self.quality.min(STEPS.len() - 1)]),
project,
};
let dirs = studio.host.dirs.clone();
@@ -257,29 +353,102 @@ impl SpeechPane {
}
/// What the sheet's script would take to say, for the line under it.
- fn estimate(&self) -> String {
+ fn estimate(&self, studio: &Studio) -> String {
let chars = self.text.trim().chars().count();
if chars == 0 {
return t("Nothing to read yet");
}
- let seconds = chars as f32 / CHARS_PER_SECOND / PACES[self.pace.min(2)];
+ let seconds = chars as f32 / CHARS_PER_SECOND / self.rate();
let whole = seconds.round() as i32;
- let voice = self
- .offered()
- .get(self.voice)
- .map(|speaker| voice_label(&speaker.name).0)
- .unwrap_or_default();
+ let voice = if self.use_sample {
+ self.sample_media(studio)
+ .map(|sample| sample.name.clone())
+ .unwrap_or_default()
+ } else {
+ self.offered()
+ .get(self.voice)
+ .map(|speaker| voice_label(&speaker.name).0)
+ .unwrap_or_default()
+ };
tf(
"About {0}:{1} in {2} · {3} characters",
&[&(whole / 60), &format!("{:02}", whole % 60), &voice, &chars],
)
}
+ /// Whether the chosen model reads at a speed and with a pause of the
+ /// sheet's choosing. Chatterbox does neither: it reads at its own
+ /// pace, and the sheet does not offer what the engine would ignore.
+ pub fn reads_at_a_pace(&self) -> bool {
+ self.family != Family::Chatterbox
+ }
+
+ /// The rate the request carries: the slider's, or 1 for a model that
+ /// reads at its own.
+ fn rate(&self) -> f32 {
+ if self.reads_at_a_pace() {
+ self.speed
+ } else {
+ 1.0
+ }
+ }
+
+ /// The id under which the chosen model reads in a recording of the
+ /// sheet's choosing; None for a model that reads only with its own
+ /// speakers.
+ fn sample_voice(&self) -> Option<i32> {
+ match self.family {
+ Family::Kokoro => None,
+ Family::Pocket => Some(POCKET_CLONE),
+ Family::Chatterbox => Some(CHATTERBOX_CLONE),
+ }
+ }
+
+ /// The recordings a voice can be taken from: everything in the bin with
+ /// sound in it, in the bin's order.
+ pub fn samples<'a>(&self, studio: &'a Studio) -> Vec<&'a MediaItem> {
+ studio
+ .project()
+ .media
+ .iter()
+ .filter(|item| item.has_audio)
+ .collect()
+ }
+
+ /// The chosen sample.
+ fn sample_media<'a>(&self, studio: &'a Studio) -> Option<&'a MediaItem> {
+ self.samples(studio).get(self.sample).copied()
+ }
+
+ /// The row to open the sample list on: the selected clip's recording,
+ /// when one clip with sound in it is selected, else the first.
+ fn default_sample(&self, studio: &Studio) -> usize {
+ studio
+ .sole_selection()
+ .and_then(|id| studio.clip(&id))
+ .filter(|clip| clip.kind == ClipKind::Video || clip.kind == ClipKind::Audio)
+ .and_then(|clip| {
+ self.samples(studio)
+ .iter()
+ .position(|item| item.id == clip.media_id)
+ })
+ .unwrap_or(0)
+ }
+
+ /// How far into a recording of `duration` the voice is taken from: what
+ /// the sheet asked for, held short of the end so there is always at
+ /// least a second of sound left to listen to.
+ fn reference_offset(&self, duration: f64) -> f64 {
+ self.reference_start.clamp(0.0, (duration - 1.0).max(0.0))
+ }
+
/// The voices the chosen model reads with, in table order.
pub fn offered(&self) -> Vec<&VoiceInfo> {
+ // The family's clone id is not a voice to pick from a list: it is
+ // what a sample is sent under, and the sample switch offers it.
self.speakers
.iter()
- .filter(|speaker| speaker.family == self.family)
+ .filter(|speaker| speaker.family == self.family && !is_clone(speaker.id))
.collect()
}
@@ -290,7 +459,7 @@ impl SpeechPane {
let default = match self.family {
Family::Kokoro => DEFAULT_VOICE,
Family::Pocket => DEFAULT_POCKET_VOICE,
- Family::Chatterbox => CHATTERBOX_CLONE,
+ Family::Chatterbox => DEFAULT_CHATTERBOX_VOICE,
};
wanted
.and_then(|id| offered.iter().position(|voice| voice.id == id))
@@ -315,14 +484,80 @@ impl SpeechPane {
.collect()
}
+ /// The model list's second lines: what each model is for, in a few
+ /// words, since which one is chosen decides everything under it.
+ pub fn model_detail_rows(&self, studio: &Studio) -> Vec<SharedString> {
+ installed(&studio.settings.voices)
+ .iter()
+ .map(|model| match family_of(&model.id) {
+ Family::Kokoro => t("Built-in speakers · fast"),
+ Family::Pocket => t("Any voice, from a recording"),
+ Family::Chatterbox => t("Studio voice · slow"),
+ })
+ .map(SharedString::from)
+ .collect()
+ }
+
+ /// The sample list's rows: each recording's name.
+ pub fn sample_rows(&self, studio: &Studio) -> Vec<SharedString> {
+ self.samples(studio)
+ .iter()
+ .map(|item| item.name.as_str().into())
+ .collect()
+ }
+
+ /// The sample list's pictures: each recording's envelope as path
+ /// commands, the same drawing the bin's card and the timeline's clip
+ /// make of it, or empty while the peaks have not been read yet.
+ pub fn sample_wave_rows(&self, studio: &Studio) -> Vec<SharedString> {
+ self.samples(studio)
+ .iter()
+ .map(|item| match studio.peaks.get(&item.id) {
+ Some(peaks) => {
+ wave_path(peaks, 0.0, item.duration.unwrap_or(0.0) as f32, 64, 0.75).into()
+ }
+ None => SharedString::new(),
+ })
+ .collect()
+ }
+
+ /// The sample list's second column: how long each recording is.
+ pub fn sample_detail_rows(&self, studio: &Studio) -> Vec<SharedString> {
+ self.samples(studio)
+ .iter()
+ .map(|item| {
+ let whole = item.duration.unwrap_or(0.0).round() as i64;
+ format!("{}:{:02}", whole / 60, whole % 60).into()
+ })
+ .collect()
+ }
+
/// The sheet as Slint shows it.
pub fn data(&self, studio: &Studio) -> SpeechSheetData {
+ let sample = self.sample_media(studio);
+ let length = sample.and_then(|item| item.duration).unwrap_or(0.0);
SpeechSheetData {
open: self.open,
text: self.text.as_str().into(),
voice: self.voice as i32,
model: self.model as i32,
- pace: self.pace as i32,
+ family: match self.family {
+ Family::Kokoro => 0,
+ Family::Pocket => 1,
+ Family::Chatterbox => 2,
+ },
+ speed: self.speed,
+ quality: self.quality as i32,
+ pauses: self.pauses,
+ samples_offered: self.sample_voice().is_some(),
+ use_sample: self.use_sample,
+ sample: self.sample as i32,
+ sample_name: sample
+ .map(|item| item.name.as_str())
+ .unwrap_or_default()
+ .into(),
+ reference_length: length as f32,
+ reference_start: self.reference_offset(length) as f32,
running: self.running,
progress: self.progress,
ready: !installed(&studio.settings.voices).is_empty(),
@@ -332,12 +567,37 @@ impl SpeechPane {
"at the playhead"
}
.into(),
- estimate: self.estimate().into(),
+ estimate: self.estimate(studio).into(),
message: self.message.as_str().into(),
}
}
}
+/// `tag` written into `text` at byte offset `at` - the caret, as the script
+/// box reports it - with a space on whichever side is against a word, so
+/// `[laugh]` lands as a word of its own and never as the tail of one. An
+/// offset inside a character steps back to the character's start; one past
+/// the end is the end. A trailing space is always written when the tag is
+/// not already followed by one, so the caller can put the caret after it.
+pub fn insert_tag(text: &str, tag: &str, at: usize) -> String {
+ let mut at = at.min(text.len());
+ while !text.is_char_boundary(at) {
+ at -= 1;
+ }
+ let (before, after) = text.split_at(at);
+ let lead = if before.is_empty() || before.ends_with(char::is_whitespace) {
+ ""
+ } else {
+ " "
+ };
+ let trail = if after.starts_with(char::is_whitespace) {
+ ""
+ } else {
+ " "
+ };
+ format!("{before}{lead}{tag}{trail}{after}")
+}
+
/// "af_heart" as a person would say it: the name, and the accent and
/// gender its prefix encodes. A Pocket voice says where it comes from.
fn voice_label(name: &str) -> (String, String) {
@@ -388,7 +648,27 @@ fn voice_label(name: &str) -> (String, String) {
#[cfg(test)]
mod tests {
- use super::voice_label;
+ use super::{insert_tag, voice_label};
+
+ /// A tag lands as a word of its own wherever the caret is: spaced off a
+ /// word on either side, not doubled against a space already there, and
+ /// snapped to a character when the caret is reported inside one.
+ #[test]
+ fn a_tag_lands_as_its_own_word() {
+ assert_eq!(insert_tag("", "[laugh]", 0), "[laugh] ");
+ assert_eq!(insert_tag("hello", "[laugh]", 5), "hello [laugh] ");
+ assert_eq!(insert_tag("hello world", "[sigh]", 5), "hello [sigh] world");
+ assert_eq!(insert_tag("hello world", "[sigh]", 6), "hello [sigh] world");
+ assert_eq!(
+ insert_tag("hello world", "[sigh]", 3),
+ "hel [sigh] lo world"
+ );
+ assert_eq!(insert_tag("hi", "[chuckle]", 0), "[chuckle] hi");
+ // Past the end is the end; inside "é" (two bytes) steps back to it.
+ assert_eq!(insert_tag("hi", "[laugh]", 40), "hi [laugh] ");
+ assert_eq!(insert_tag("café", "[laugh]", 4), "caf [laugh] é");
+ assert_eq!(insert_tag("café ", "[laugh]", 6), "café [laugh] ");
+ }
#[test]
fn a_voice_name_reads_as_a_person_would_say_it() {
diff --git a/src/crates/concat/src/studio.rs b/src/crates/concat/src/studio.rs
index 5b9cc278..f5285bef 100644
--- a/src/crates/concat/src/studio.rs
+++ b/src/crates/concat/src/studio.rs
@@ -495,8 +495,12 @@ pub struct Models {
/// The sheets' option lists: what is installed, and who can speak.
pub caption_models: Rc<VecModel<SharedString>>,
pub speech_models: Rc<VecModel<SharedString>>,
+ pub speech_model_details: Rc<VecModel<SharedString>>,
pub speakers: Rc<VecModel<SharedString>>,
pub speaker_details: Rc<VecModel<SharedString>>,
+ pub speech_samples: Rc<VecModel<SharedString>>,
+ pub speech_sample_waves: Rc<VecModel<SharedString>>,
+ pub speech_sample_details: Rc<VecModel<SharedString>>,
pub transcribers: Rc<VecModel<ModelData>>,
pub voices: Rc<VecModel<ModelData>>,
pub seats: Rc<VecModel<SeatBox>>,
@@ -536,8 +540,12 @@ impl Models {
bar: Rc::new(VecModel::default()),
caption_models: Rc::new(VecModel::default()),
speech_models: Rc::new(VecModel::default()),
+ speech_model_details: Rc::new(VecModel::default()),
speakers: Rc::new(VecModel::default()),
speaker_details: Rc::new(VecModel::default()),
+ speech_samples: Rc::new(VecModel::default()),
+ speech_sample_waves: Rc::new(VecModel::default()),
+ speech_sample_details: Rc::new(VecModel::default()),
transcribers: Rc::new(VecModel::default()),
voices: Rc::new(VecModel::default()),
seats: Rc::new(VecModel::default()),
@@ -6520,8 +6528,21 @@ impl Studio {
.map(|model| SharedString::from(model.name.as_str()))
.collect(),
);
+ sync(
+ &models.speech_model_details,
+ self.speech.model_detail_rows(self),
+ );
sync(&models.speakers, self.speech.speaker_rows());
sync(&models.speaker_details, self.speech.speaker_detail_rows());
+ sync(&models.speech_samples, self.speech.sample_rows(self));
+ sync(
+ &models.speech_sample_waves,
+ self.speech.sample_wave_rows(self),
+ );
+ sync(
+ &models.speech_sample_details,
+ self.speech.sample_detail_rows(self),
+ );
app.set_speech(self.speech.data(self));
let bar = self.menu_bar();
diff --git a/src/crates/concat/ui/app.slint b/src/crates/concat/ui/app.slint
index 1a9e5a4f..bfb8f18a 100644
--- a/src/crates/concat/ui/app.slint
+++ b/src/crates/concat/ui/app.slint
@@ -247,13 +247,26 @@ export component App inherits Window {
in property <SpeechSheetData> speech;
in property <[string]> speech-models;
+ in property <[string]> speech-model-details;
in property <[string]> speech-voices;
in property <[string]> speech-voice-details;
+ /// The recordings in the bin a voice can be taken from: their names,
+ /// their waveforms as path commands, and their lengths.
+ in property <[string]> speech-samples;
+ in property <[string]> speech-sample-waves;
+ in property <[string]> speech-sample-details;
callback speech-closed();
callback speech-text-edited(text: string);
callback speech-voice-changed(index: int);
callback speech-model-changed(index: int);
- callback speech-pace-changed(index: int);
+ callback speech-speed-changed(value: float);
+ callback speech-quality-changed(index: int);
+ callback speech-pauses-changed(value: float);
+ callback speech-reference-start-changed(seconds: float);
+ callback speech-use-sample-changed(on: bool);
+ callback speech-sample-changed(index: int);
+ /// See SpeechDialog.insert-tag: a string cut Rust does for the sheet.
+ pure callback speech-insert-tag(text: string, tag: string, at: int) -> string;
callback speech-begin();
callback speech-cancel();
@@ -597,14 +610,24 @@ export component App inherits Window {
height: root.height;
data: root.speech;
models: root.speech-models;
+ model-details: root.speech-model-details;
voices: root.speech-voices;
voice-details: root.speech-voice-details;
+ samples: root.speech-samples;
+ sample-waves: root.speech-sample-waves;
+ sample-details: root.speech-sample-details;
closed => { root.speech-closed(); }
text-edited(text) => { root.speech-text-edited(text); }
voice-changed(index) => { root.speech-voice-changed(index); }
model-changed(index) => { root.speech-model-changed(index); }
- pace-changed(index) => { root.speech-pace-changed(index); }
+ speed-changed(value) => { root.speech-speed-changed(value); }
+ quality-changed(index) => { root.speech-quality-changed(index); }
+ pauses-changed(value) => { root.speech-pauses-changed(value); }
+ reference-start-changed(seconds) => { root.speech-reference-start-changed(seconds); }
+ use-sample-changed(on) => { root.speech-use-sample-changed(on); }
+ sample-changed(index) => { root.speech-sample-changed(index); }
+ insert-tag(text, tag, at) => { return root.speech-insert-tag(text, tag, at); }
begin => { root.speech-begin(); }
cancel => { root.speech-cancel(); }
}
diff --git a/src/crates/concat/ui/dialogs/speech.slint b/src/crates/concat/ui/dialogs/speech.slint
index 1b808da9..224575dd 100644
--- a/src/crates/concat/ui/dialogs/speech.slint
+++ b/src/crates/concat/ui/dialogs/speech.slint
@@ -7,14 +7,25 @@ import { Glyph, Icon } from "../icons.slint";
import { Button, CopyChip, DialogRow, DialogSection, Modal, Progress } from "../primitives/modal.slint";
import { SegmentedControl } from "../primitives/segmented-control.slint";
import { Select } from "../primitives/select.slint";
+import { Rail } from "../inspector/controls.slint";
+import { Wave } from "../primitives/wave.slint";
+import { Fmt } from "../util.slint";
// Text to speech: what the tray's Speak tool and File › Text to speech open.
//
// A text box, a voice and a button is enough to send a job and not enough to
-// get the read you wanted, so the three things that change a read are on the
-// surface: which voice, which model, and the pace. Opened from a title the
-// script is the title's words; opened from the menu it is whatever you type.
-// While it runs the sheet is the progress report, with a clock beside it.
+// get the read you wanted, so what changes a read is on the surface. The
+// model comes first, because it decides the rest: which voices there are to
+// choose from, and which dials the engine will read. Each model gets its own
+// controls rather than the few every model shares - Kokoro its speed and its
+// pauses, Pocket those and a quality, Chatterbox its expression tags and
+// nothing about pace, since it reads at its own. The two that read in a
+// recording can read in a sample voice instead of a named one: a section of
+// its own, switched on at its header, offering the bin's recordings by their
+// waveforms and how far into the chosen one the speaking starts. Opened
+// from a title the script is the title's words; opened from the menu it is
+// whatever you type. While it runs the sheet is the progress report, with a
+// clock beside it.
export struct SpeechSheetData {
open: bool,
@@ -23,8 +34,27 @@ export struct SpeechSheetData {
voice: int,
/// Row in the installed voice model list.
model: int,
- /// 0 Slower, 1 Natural, 2 Faster
- pace: int,
+ /// 0 Kokoro, 1 Pocket TTS, 2 Chatterbox: which controls the model has.
+ family: int,
+ /// The rate, 0.5..2; 1 is the voice's own. Kokoro and Pocket.
+ speed: float,
+ /// 0 quick, 1 balanced, 2 fine: Pocket's flow steps.
+ quality: int,
+ /// The breath between sentences, 0..1. Kokoro and Pocket.
+ pauses: float,
+ /// The model reads in a recording, so a sample voice can be offered:
+ /// Pocket and Chatterbox.
+ samples-offered: bool,
+ /// Reading in a sample from the bin rather than a named voice.
+ use-sample: bool,
+ /// Row in the sample list.
+ sample: int,
+ /// The chosen sample's name, or empty when the bin has none.
+ sample-name: string,
+ /// How long that sample is, in seconds: the room the start below has.
+ reference-length: float,
+ /// Seconds into that sample the speaking starts.
+ reference-start: float,
running: bool,
/// 0..1, as sentences complete.
progress: float,
@@ -38,17 +68,176 @@ export struct SpeechSheetData {
message: string,
}
+// A list of recordings to pick a voice from, each drawn as its waveform
+// with its name and length beside it - a voice is heard, and a row of
+// names would say nothing about which one is the clean take. The shape is
+// the Select's: a trigger showing the current row, a list under it that
+// closes on a click outside. Without the Select's keyboard, since the
+// rows are pictures and there are seldom more than a handful.
+component SamplePicker inherits Rectangle {
+ in property <[string]> names;
+ /// Each recording's envelope as path commands; shorter than `names`
+ /// is fine, and an empty one draws no wave.
+ in property <[string]> waves;
+ in property <[string]> details;
+ in-out property <int> current;
+
+ callback changed(index: int);
+
+ property <length> row-h: 36px;
+ property <length> content-h: root.names.length * root.row-h + 8px;
+ property <length> list-h: Math.min(4 * root.row-h + 8px, root.content-h);
+ property <bool> has-current: root.current >= 0 && root.current < root.names.length;
+
+ height: 34px;
+ horizontal-stretch: 1;
+ border-radius: Theme.r;
+ background: (list.is-open || trigger.has-hover) ? Theme.field-hover : Theme.field;
+
+ animate background { duration: 120ms; easing: cubic-bezier(0.4, 0, 0.2, 1); }
+
+ HorizontalLayout {
+ padding-left: 8px;
+ padding-right: 6px;
+ spacing: 8px;
+
+ Rectangle {
+ width: 96px;
+ if root.has-current && root.current < root.waves.length: Wave {
+ x: 0;
+ y: 6px;
+ width: parent.width;
+ height: parent.height - 12px;
+ commands: root.waves[root.current];
+ tint: Theme.kind-audio;
+ }
+ }
+
+ Text {
+ horizontal-stretch: 1;
+ text: root.has-current ? root.names[root.current] : I18n.t("Pick a recording");
+ color: root.has-current ? Theme.fg : Theme.fg-dim;
+ font-size: Theme.fs;
+ vertical-alignment: center;
+ overflow: elide;
+ }
+
+ VerticalLayout {
+ alignment: center;
+ Icon { glyph: Glyph.chevron-down; size: 12px; tint: Theme.fg-muted; }
+ }
+ }
+
+ trigger := TouchArea {
+ mouse-cursor: pointer;
+ clicked => { list.show(); }
+ }
+
+ list := PopupWindow {
+ x: 0;
+ y: root.height + 4px;
+ width: root.width;
+ height: root.list-h;
+ close-policy: close-on-click-outside;
+
+ Rectangle {
+ background: Theme.raised;
+ border-width: 1px;
+ border-color: Theme.line-strong;
+ border-radius: Theme.r;
+ clip: true;
+
+ Flickable {
+ viewport-height: root.content-h;
+
+ VerticalLayout {
+ padding: 4px;
+ alignment: start;
+
+ for name[index] in root.names: Rectangle {
+ height: root.row-h;
+ border-radius: Theme.r-sm;
+ background: index == root.current ? Theme.accent-soft
+ : pick.has-hover ? Theme.field
+ : Colors.transparent;
+
+ HorizontalLayout {
+ padding-left: 6px;
+ padding-right: 8px;
+ spacing: 8px;
+
+ Rectangle {
+ width: 96px;
+ if index < root.waves.length: Wave {
+ x: 0;
+ y: 6px;
+ width: parent.width;
+ height: parent.height - 12px;
+ commands: root.waves[index];
+ tint: Theme.kind-audio;
+ }
+ }
+
+ Text {
+ horizontal-stretch: 1;
+ text: name;
+ color: index == root.current ? Theme.fg : Theme.fg-muted;
+ font-size: Theme.fs;
+ vertical-alignment: center;
+ overflow: elide;
+ }
+
+ if index < root.details.length: Text {
+ text: root.details[index];
+ color: Theme.fg-dim;
+ font-family: Theme.font-technical;
+ font-size: Theme.fs-sm;
+ vertical-alignment: center;
+ }
+ }
+
+ pick := TouchArea {
+ mouse-cursor: pointer;
+ clicked => {
+ root.current = index;
+ root.changed(index);
+ list.close();
+ }
+ }
+ }
+ }
+ }
+ }
+ }
+}
+
export component SpeechDialog inherits Modal {
in property <SpeechSheetData> data;
in property <[string]> models;
+ /// "Built-in speakers · fast", beside each model.
+ in property <[string]> model-details;
in property <[string]> voices;
/// "American · female", beside each voice.
in property <[string]> voice-details;
+ /// The bin's recordings a voice can be taken from: their names, their
+ /// envelopes as path commands, and their lengths.
+ in property <[string]> samples;
+ in property <[string]> sample-waves;
+ in property <[string]> sample-details;
callback text-edited(text: string);
callback voice-changed(index: int);
callback model-changed(index: int);
- callback pace-changed(index: int);
+ callback speed-changed(value: float);
+ callback quality-changed(index: int);
+ callback pauses-changed(value: float);
+ callback reference-start-changed(seconds: float);
+ callback use-sample-changed(on: bool);
+ callback sample-changed(index: int);
+ /// `tag` written into `text` at byte offset `at`, spaced as a word of
+ /// its own. Rust does the writing: the expression language has no way
+ /// to cut a string at an offset.
+ pure callback insert-tag(text: string, tag: string, at: int) -> string;
callback begin();
callback cancel();
@@ -68,12 +257,33 @@ export component SpeechDialog inherits Modal {
/// The script has the caret. Kept up here because the box lives under
/// an `if`, and nothing outside one can reach into it.
property <bool> typing;
+ // The sliders' own copies, seeded the same way: a Rail writes to the
+ // property it is bound to, which drops the binding, so each is reseeded
+ // from the data whenever it changes - and Rust echoes every change
+ // back, so the reseed is the value the slider just set.
+ property <float> speed-v;
+ property <float> pauses-v;
+ property <float> reference-start-v;
changed data => {
if (!root.typing) {
root.draft = root.data.text;
}
+ root.speed-v = root.data.speed;
+ root.pauses-v = root.data.pauses;
+ root.reference-start-v = root.data.reference-start;
}
- init => { root.draft = root.data.text; }
+ init => {
+ root.draft = root.data.text;
+ root.speed-v = root.data.speed;
+ root.pauses-v = root.data.pauses;
+ root.reference-start-v = root.data.reference-start;
+ }
+ /// A sample is wanted and the bin has one to give; a read without one
+ /// would fail on its first step.
+ property <bool> has-reference: !root.data.use-sample || root.data.sample-name != "";
+ /// There is a voice to read with: a sample, or a named one on the list.
+ /// Chatterbox names none until Pocket's recordings are on disk.
+ property <bool> has-voice: root.data.use-sample || root.voices.length > 0;
/// Seconds since the run began; the reference beside the bar.
property <int> elapsed;
@@ -135,6 +345,35 @@ export component SpeechDialog inherits Modal {
}
}
+ // Chatterbox alone reads these as what they say, so the row is
+ // its alone. Each lands at the caret as a word of its own, and
+ // the caret moves past it, so a laugh can be dropped mid-line
+ // without leaving the keyboard for long.
+ if root.data.family == 2: HorizontalLayout {
+ spacing: 6px;
+ alignment: start;
+
+ Text {
+ text: I18n.t("Expression");
+ color: Theme.fg-muted;
+ font-size: Theme.fs;
+ vertical-alignment: center;
+ }
+
+ for tag in ["[laugh]", "[chuckle]", "[sigh]"]: Button {
+ label: tag;
+ ghost: true;
+ clicked => {
+ let at = script.cursor-position-byte-offset;
+ root.draft = root.insert-tag(root.draft, tag, at);
+ root.text-edited(root.draft);
+ script.focus();
+ script.set-selection-offsets(at + tag.character-count + 1,
+ at + tag.character-count + 1);
+ }
+ }
+ }
+
Text {
text: root.data.estimate;
color: Theme.fg-dim;
@@ -146,33 +385,111 @@ export component SpeechDialog inherits Modal {
if !root.running: DialogSection {
title: I18n.t("Voice");
+ // The model first: it decides which voices there are and which
+ // of the dials below are shown.
DialogRow {
+ label: I18n.t("Model");
+ Select {
+ options: root.models;
+ details: root.model-details;
+ current: root.data.model;
+ placeholder: I18n.t("None installed");
+ enabled: root.models.length > 0;
+ changed(index) => { root.model-changed(index); }
+ }
+ }
+
+ // The named voices, while a sample is not standing in for them.
+ if !root.data.use-sample: DialogRow {
label: I18n.t("Voice");
Select {
options: root.voices;
details: root.voice-details;
current: root.data.voice;
+ placeholder: I18n.t("None");
+ enabled: root.voices.length > 0;
changed(index) => { root.voice-changed(index); }
}
}
- DialogRow {
- label: I18n.t("Model");
- Select {
- options: root.models;
- current: root.data.model;
- placeholder: I18n.t("None installed");
- enabled: root.models.length > 0;
- changed(index) => { root.model-changed(index); }
+ // Chatterbox ships no recording of anyone's, so its named voices
+ // are Pocket's two, borrowed while Pocket is on disk. With
+ // neither there is nothing to read with but a sample.
+ if root.data.family == 2 && root.voices.length == 0 && !root.data.use-sample: Text {
+ text: I18n.t("Chatterbox has no voice of its own. Turn on Use sample voice below, or download Pocket TTS in Settings › Speech to borrow its two recordings.");
+ color: Theme.danger;
+ font-size: Theme.fs-sm;
+ wrap: word-wrap;
+ }
+
+ // What this model is, in a line, so the dials under it are not
+ // a surprise: a model that is missing one has never had it.
+ if root.data.ready: Text {
+ text: root.data.family == 0
+ ? I18n.t("Speakers built into the model, in English and Chinese. Reads at the speed and with the pauses set below.")
+ : root.data.family == 1
+ ? I18n.t("Reads in any voice from a few seconds of a recording: one it ships with, or a sample from the bin.")
+ : I18n.t("The closest to a studio voice. Reads at its own pace, in one of Pocket's two recordings or in any sample from the bin, and reads [laugh], [chuckle] and [sigh] as what they say.");
+ color: Theme.fg-dim;
+ font-size: Theme.fs-sm;
+ wrap: word-wrap;
+ }
+
+ // Speed and pauses are sherpa's dials, and Chatterbox is not
+ // sherpa's: it reads at its own pace, and a slider it would
+ // ignore is a lie.
+ if root.data.family != 2: DialogRow {
+ label: I18n.t("Speed");
+ Rail {
+ value <=> root.speed-v;
+ minimum: 0.5;
+ maximum: 2;
+ step: 0.05;
+ default-value: 1;
+ changed(v) => { root.speed-changed(v); }
+ }
+ Text {
+ width: 52px;
+ text: "\{Fmt.round-to(root.speed-v, 2)}×";
+ color: Theme.fg-muted;
+ font-family: Theme.font-technical;
+ font-size: Theme.fs-sm;
+ vertical-alignment: center;
+ horizontal-alignment: right;
}
}
- DialogRow {
- label: I18n.t("Pace");
+ if root.data.family != 2: DialogRow {
+ label: I18n.t("Pauses");
+ Rail {
+ value <=> root.pauses-v;
+ minimum: 0;
+ maximum: 1;
+ step: 0.05;
+ default-value: 0.2;
+ changed(v) => { root.pauses-changed(v); }
+ }
+ Text {
+ width: 52px;
+ text: root.pauses-v < 0.1 ? I18n.t("none")
+ : root.pauses-v < 0.35 ? I18n.t("natural")
+ : root.pauses-v < 0.7 ? I18n.t("long")
+ : I18n.t("longer");
+ color: Theme.fg-muted;
+ font-family: Theme.font-technical;
+ font-size: Theme.fs-sm;
+ vertical-alignment: center;
+ horizontal-alignment: right;
+ }
+ }
+
+ // Pocket's flow steps: fewer is quicker, more is cleaner.
+ if root.data.family == 1: DialogRow {
+ label: I18n.t("Quality");
SegmentedControl {
- options: [I18n.t("Slower"), I18n.t("Natural"), I18n.t("Faster")];
- current: root.data.pace;
- changed(index) => { root.pace-changed(index); }
+ options: [I18n.t("Quick"), I18n.t("Balanced"), I18n.t("Fine")];
+ current: root.data.quality;
+ changed(index) => { root.quality-changed(index); }
}
}
@@ -184,6 +501,60 @@ export component SpeechDialog inherits Modal {
}
}
+ // A sample voice: a recording from the bin in place of a named
+ // voice, for the two models that read in one. The switch is at the
+ // section's head, and the body folds away while it is off. The
+ // recordings are offered by their waveforms, since a voice is heard
+ // and not read, and the start slider says where in the recording
+ // the model listens from.
+ if !root.running && root.data.samples-offered: DialogSection {
+ title: I18n.t("Use sample voice");
+ switchable: true;
+ on: root.data.use-sample;
+ open: root.data.use-sample;
+ switched(on) => { root.use-sample-changed(on); }
+ hint: I18n.t("Reads in the voice heard in a recording from the bin. The first ten seconds from the point chosen below are what the model listens to.");
+
+ if root.samples.length == 0: Text {
+ text: I18n.t("Import a clip with a voice in it first.");
+ color: Theme.fg-muted;
+ font-size: Theme.fs-sm;
+ wrap: word-wrap;
+ }
+
+ if root.samples.length > 0: DialogRow {
+ label: I18n.t("Sample");
+ SamplePicker {
+ names: root.samples;
+ waves: root.sample-waves;
+ details: root.sample-details;
+ current: root.data.sample;
+ changed(index) => { root.sample-changed(index); }
+ }
+ }
+
+ if root.samples.length > 0: DialogRow {
+ label: I18n.t("Voice from");
+ Rail {
+ value <=> root.reference-start-v;
+ minimum: 0;
+ maximum: Math.max(0, root.data.reference-length - 1);
+ step: 0.5;
+ default-value: 0;
+ changed(v) => { root.reference-start-changed(v); }
+ }
+ Text {
+ width: 52px;
+ text: I18n.t1("{0} s in", "\{Fmt.round-to(root.reference-start-v, 1)}");
+ color: Theme.fg-muted;
+ font-family: Theme.font-technical;
+ font-size: Theme.fs-sm;
+ vertical-alignment: center;
+ horizontal-alignment: right;
+ }
+ }
+ }
+
Rectangle {
height: 1px;
background: Theme.line;
@@ -269,7 +640,7 @@ export component SpeechDialog inherits Modal {
large: true;
glyph: Glyph.volume;
label: I18n.t("Generate speech");
- enabled: root.data.ready && root.data.text != "";
+ enabled: root.data.ready && root.data.text != "" && root.has-reference && root.has-voice;
clicked => { root.begin(); }
}
}
diff --git a/src/crates/concat/ui/primitives/modal.slint b/src/crates/concat/ui/primitives/modal.slint
index 3c0e1abc..60a045a1 100644
--- a/src/crates/concat/ui/primitives/modal.slint
+++ b/src/crates/concat/ui/primitives/modal.slint
@@ -344,6 +344,15 @@ export component DialogSection inherits VerticalLayout {
/// a question mark beside the title: read on hover, out of the way
/// otherwise.
in property <string> hint;
+ /// A switch at the header's far end, for a group that is a feature to
+ /// turn on rather than a set of settings that are always there. The
+ /// switch draws `on` and reports a press through `switched`; it keeps
+ /// no state of its own, so the caller's data stays the one truth and
+ /// a reopened sheet cannot disagree with it. Bind `open` to the same
+ /// value to fold the body away while it is off.
+ in property <bool> switchable;
+ in property <bool> on;
+ callback switched(on: bool);
if !root.first: Rectangle {
height: 1px;
@@ -422,6 +431,37 @@ export component DialogSection inherits VerticalLayout {
vertical-alignment: center;
overflow: elide;
}
+
+ // The same pill as the inspector's Switch, drawn from `on`
+ // rather than holding a checked state of its own.
+ if root.switchable: VerticalLayout {
+ alignment: center;
+
+ Rectangle {
+ width: 30px;
+ height: 17px;
+ border-radius: self.height / 2;
+ background: root.on ? Theme.accent : Theme.field-active;
+
+ animate background { duration: 120ms; easing: cubic-bezier(0.4, 0, 0.2, 1); }
+
+ Rectangle {
+ x: root.on ? parent.width - self.width - 2px : 2px;
+ y: (parent.height - self.height) / 2;
+ width: 13px;
+ height: 13px;
+ border-radius: self.width / 2;
+ background: Theme.on-accent;
+
+ animate x { duration: 120ms; easing: cubic-bezier(0.4, 0, 0.2, 1); }
+ }
+
+ TouchArea {
+ mouse-cursor: pointer;
+ clicked => { root.switched(!root.on); }
+ }
+ }
+ }
}
}