Skip to content

Open source Grok Bot alternative, Bring your own agent and subscription: message Claude Code, Codex, Gemini CLI, Cursor, Copilot and 40+ AI agents as bots on your own computer, from iPhone, Mac, Windows, Linux or terminal. Agent-agnostic and free. Also a Muse and Dots alternative. Built in Rust.

feat(desktop): choose the language for on-device voice calls

leepokai25a7f2bf15c7
diff --git a/apps/desktop/native/speech-macos/Sources/main.swift b/apps/desktop/native/speech-macos/Sources/main.swift
index 5b610d51..3cb1827e 100644
--- a/apps/desktop/native/speech-macos/Sources/main.swift
+++ b/apps/desktop/native/speech-macos/Sources/main.swift
@@ -5,7 +5,7 @@ import Speech
// codync-speech: transcribes the microphone on device with SFSpeechRecognizer and prints one
// JSON event per line: {"type":"started"}, {"type":"partial"|"final","text":…},
// {"type":"level","value":0…1}, {"type":"error","message":…}. It runs until stdin closes or
-// SIGTERM. The desktop app (src/main/speech.ts) owns endpointing: it stops this process when
+// SIGTERM. `--locales` prints the languages it can transcribe (a JSON array) and exits. The desktop app (src/main/speech.ts) owns endpointing: it stops this process when
// the speaker pauses.
@MainActor
@@ -99,6 +99,11 @@ final class Recognizer {
}
let arguments = CommandLine.arguments
+if arguments.contains("--locales") {
+ let ids = SFSpeechRecognizer.supportedLocales().map(\.identifier).sorted()
+ FileHandle.standardOutput.write((try? JSONSerialization.data(withJSONObject: ids)) ?? Data("[]".utf8))
+ exit(0)
+}
let locale = arguments.firstIndex(of: "--locale").flatMap { arguments.indices.contains($0 + 1) ? arguments[$0 + 1] : nil }
signal(SIGTERM, SIG_IGN)
diff --git a/apps/desktop/src/main/speech.ts b/apps/desktop/src/main/speech.ts
index 62e20ac0..ace32d5d 100644
--- a/apps/desktop/src/main/speech.ts
+++ b/apps/desktop/src/main/speech.ts
@@ -1,4 +1,4 @@
-import { spawn, type ChildProcessWithoutNullStreams } from 'node:child_process'
+import { execFileSync, spawn, type ChildProcessWithoutNullStreams } from 'node:child_process'
import { existsSync } from 'node:fs'
import { join } from 'node:path'
import { app, ipcMain, type WebContents } from 'electron'
@@ -58,8 +58,23 @@ function stop() {
if (proc && !proc.killed) proc.kill('SIGTERM')
}
+let supported: string[] | null = null
+
+/** The languages the recognizer can transcribe (identifiers like `zh-TW`), asked once. */
+function locales() {
+ if (supported) return supported
+ const path = helperPath()
+ try {
+ supported = path ? (JSON.parse(execFileSync(path, ['--locales'], { encoding: 'utf8', timeout: 5000 })) as string[]) : []
+ } catch {
+ supported = []
+ }
+ return supported
+}
+
export function registerSpeech() {
ipcMain.on('speech:available', (e) => (e.returnValue = process.platform === 'darwin' && helperPath() !== null))
+ ipcMain.on('speech:locales', (e) => (e.returnValue = locales()))
ipcMain.on('speech:start', (e, locale: string | null) => start(e.sender, locale))
ipcMain.on('speech:stop', () => stop())
app.on('before-quit', stop)
diff --git a/apps/desktop/src/preload/index.ts b/apps/desktop/src/preload/index.ts
index 8b79014a..0dd51f49 100644
--- a/apps/desktop/src/preload/index.ts
+++ b/apps/desktop/src/preload/index.ts
@@ -57,6 +57,7 @@ const bridge: CodyncBridge = {
},
speech: {
available: ipcRenderer.sendSync('speech:available') as boolean,
+ locales: () => ipcRenderer.sendSync('speech:locales') as string[],
start: (locale) => ipcRenderer.send('speech:start', locale),
stop: () => ipcRenderer.send('speech:stop'),
onEvent: (cb) => listen('speech:event', cb),
diff --git a/apps/desktop/src/renderer/views/call/OnDeviceEngine.ts b/apps/desktop/src/renderer/views/call/OnDeviceEngine.ts
index e3adc2db..b7e5c34c 100644
--- a/apps/desktop/src/renderer/views/call/OnDeviceEngine.ts
+++ b/apps/desktop/src/renderer/views/call/OnDeviceEngine.ts
@@ -117,7 +117,7 @@ export class OnDeviceEngine extends VoiceEngine {
if (this.muted || this.ended) return
this.recognizing = true
this.started = false
- window.codync.speech.start(null)
+ window.codync.speech.start(VoiceSettings.language() || null)
}
private stopListening() {
diff --git a/apps/desktop/src/renderer/views/call/voice.ts b/apps/desktop/src/renderer/views/call/voice.ts
index 983e7700..6bada123 100644
--- a/apps/desktop/src/renderer/views/call/voice.ts
+++ b/apps/desktop/src/renderer/views/call/voice.ts
@@ -5,6 +5,7 @@ import {
chosenVoice,
engine,
isVoiceProvider,
+ language,
mode,
pause,
rate,
@@ -44,6 +45,8 @@ export const VoiceSettings = {
pause: () => (pause.get() > 0 ? pause.get() : 1.5),
/** AVSpeechUtterance's rate (0…1, 0.5 normal), as the Mac stores it. */
rate: () => rate.get(),
+ /** The on-device recognizer's language; '' is the computer's. */
+ language: () => language.get(),
}
/** This month's realtime minutes, counted on this device. */
diff --git a/apps/desktop/src/renderer/views/settings/VoiceChatSettingsView.tsx b/apps/desktop/src/renderer/views/settings/VoiceChatSettingsView.tsx
index 98fe7cba..041d7c1e 100644
--- a/apps/desktop/src/renderer/views/settings/VoiceChatSettingsView.tsx
+++ b/apps/desktop/src/renderer/views/settings/VoiceChatSettingsView.tsx
@@ -1,5 +1,5 @@
import { onDeviceUnavailableReason } from '../call/voice'
-import { useEffect, useRef, useState } from 'react'
+import { useEffect, useMemo, useRef, useState } from 'react'
import type { HostClient } from '../../client/host-client'
import { Button, CardForm, CardSection, ChoicePicker, IconButton, Spinner, ValueRow } from '../../components/Controls'
import { Icon } from '../../components/Icon'
@@ -19,6 +19,8 @@ import type { VoiceModels, VoiceProvider } from './voice-settings'
export function VoiceChatSettingsView({ showsHeader = true }: { showsHeader?: boolean }) {
const [engine, setEngine] = usePref(voice.engine)
const [rate, setRate] = usePref(voice.rate)
+ const [language, setLanguage] = usePref(voice.language)
+ const languages = useMemo(() => voice.languageOptions(window.codync.speech.locales()), [])
const provider = voice.isVoiceProvider(engine) ? engine : null
const [mode, setMode] = usePref(voice.mode(provider ?? 'openai'))
const footer = provider
@@ -48,6 +50,9 @@ export function VoiceChatSettingsView({ showsHeader = true }: { showsHeader?: bo
...(mode === 'speech' ? [<PauseRow key="pause" />] : []),
]
: [
+ <ValueRow key="language" label="Language" detail="The language you speak in calls.">
+ <ChoicePicker selection={language} options={languages} onChange={setLanguage} />
+ </ValueRow>,
<ValueRow key="rate" label="Reading speed">
<ChoicePicker
selection={rate}
diff --git a/apps/desktop/src/renderer/views/settings/voice-settings.ts b/apps/desktop/src/renderer/views/settings/voice-settings.ts
index c96bbf4f..fb7937d6 100644
--- a/apps/desktop/src/renderer/views/settings/voice-settings.ts
+++ b/apps/desktop/src/renderer/views/settings/voice-settings.ts
@@ -61,6 +61,24 @@ export const DEFAULT_RATE = 0.5
export const rate = pref('callRate', DEFAULT_RATE)
/** Seconds of silence that end what you say. */
export const pause = pref('callPause', 1.5)
+/** On-device recognition language (a locale identifier); '' follows the system. */
+export const language = pref('callLanguage', '')
+
+/** Picker options: the system's language, then every language the recognizer supports, by name. */
+export function languageOptions(supported: string[]) {
+ const names = new Intl.DisplayNames([navigator.language], { type: 'language' })
+ const name = (id: string) => {
+ try {
+ return names.of(id.replace('_', '-')) ?? id
+ } catch {
+ return id
+ }
+ }
+ return [
+ { id: '', label: 'System' },
+ ...supported.map((id) => ({ id, label: name(id) })).sort((a, b) => a.label.localeCompare(b.label)),
+ ]
+}
const cache = new Map<string, Pref<unknown>>()
function keyed<T>(key: string, fallback: T): Pref<T> {
diff --git a/apps/desktop/src/shared/ipc.ts b/apps/desktop/src/shared/ipc.ts
index 2795dd1c..a2570a91 100644
--- a/apps/desktop/src/shared/ipc.ts
+++ b/apps/desktop/src/shared/ipc.ts
@@ -131,6 +131,9 @@ export interface CodyncBridge {
/** On-device speech recognition (macOS: SFSpeechRecognizer in a helper; unavailable on Linux). */
speech: {
available: boolean
+ /** The languages it can transcribe, as locale identifiers. */
+ locales(): string[]
+ /** `locale` null: the computer's language. */
start(locale: string | null): void
stop(): void
onEvent(cb: (e: SpeechEvent) => void): () => void
diff --git a/apps/ios/Kit/Sources/CodyncUI/Call/OnDeviceEngine.swift b/apps/ios/Kit/Sources/CodyncUI/Call/OnDeviceEngine.swift
index a23c10ae..f48db794 100644
--- a/apps/ios/Kit/Sources/CodyncUI/Call/OnDeviceEngine.swift
+++ b/apps/ios/Kit/Sources/CodyncUI/Call/OnDeviceEngine.swift
@@ -32,7 +32,9 @@ final class OnDeviceEngine: VoiceEngine {
}
private let send: (String) -> Void
- private let recognizer = SFSpeechRecognizer()
+ /// The language picked in the call settings, else the phone's.
+ private let recognizer = UserDefaults.standard.string(forKey: OnDeviceEngine.languageKey)
+ .flatMap { $0.isEmpty ? nil : SFSpeechRecognizer(locale: Locale(identifier: $0)) } ?? SFSpeechRecognizer()
private let engine = AVAudioEngine()
private var request: SFSpeechAudioBufferRecognitionRequest?
private var task: SFSpeechRecognitionTask?
@@ -46,6 +48,14 @@ final class OnDeviceEngine: VoiceEngine {
/// Call settings (the gear), kept across calls.
static let pauseKey = "callPause"
static let rateKey = "callRate"
+ /// On-device recognition language (a locale identifier); empty follows the system.
+ static let languageKey = "callLanguage"
+
+ /// Settings options: the system's language, then every language the recognizer supports, by name.
+ static var languageOptions: [(String, String)] {
+ let named = SFSpeechRecognizer.supportedLocales().map { ($0.identifier, Locale.current.localizedString(forIdentifier: $0.identifier) ?? $0.identifier) }
+ return [("", "System")] + named.sorted { $0.1.localizedStandardCompare($1.1) == .orderedAscending }
+ }
/// A pause this long ends what you're saying and sends it.
private var endOfUtterance: Duration {
diff --git a/apps/ios/Kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift b/apps/ios/Kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift
index 6c987d6e..d81f2ca7 100644
--- a/apps/ios/Kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift
+++ b/apps/ios/Kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift
@@ -8,6 +8,8 @@ import SwiftUI
public struct VoiceChatSettingsView: View {
@AppStorage(VoiceSettings.engineKey) private var engine = "device"
@AppStorage(OnDeviceEngine.rateKey) private var rate = Double(AVSpeechUtteranceDefaultSpeechRate)
+ @AppStorage(OnDeviceEngine.languageKey) private var language = ""
+ @State private var languages = OnDeviceEngine.languageOptions
@Environment(\.accessibilityReduceMotion) private var reduceMotion
public init() {}
@@ -24,6 +26,9 @@ public struct VoiceChatSettingsView: View {
if let provider = VoiceProvider(rawValue: engine) {
ModeRows(provider: provider).id(provider)
} else {
+ ValueRow("Language", detail: "The language you speak in calls.") {
+ ChoicePicker(selection: $language, options: languages)
+ }
ValueRow("Reading speed") {
ChoicePicker(selection: $rate, options: [
(0.42, "Slower"), (Double(AVSpeechUtteranceDefaultSpeechRate), "Normal"), (0.56, "Faster"),
diff --git a/apps/ios/WhatsNew.md b/apps/ios/WhatsNew.md
index ae620817..763379ed 100644
--- a/apps/ios/WhatsNew.md
+++ b/apps/ios/WhatsNew.md
@@ -15,6 +15,7 @@ a version without one gets a generic "Bug fixes and improvements" line. Delete o
- 進聊天室先顯示最新的訊息,往上捲再接上更早的;更新 app 之後也不用等對話重新載入。
- 回覆裡的表格、分隔線、巢狀清單和行內程式碼都會正確排版。
- 開啟 app 時會有 bot 動畫陪你等連線,畫面出現時新訊息已經同步好了。
+- 裝置上的語音通話可以選擇你說的語言(語音聊天設定 → 語言),不再只跟著系統的第一語言。
### en-US
@@ -24,3 +25,4 @@ a version without one gets a generic "Bug fixes and improvements" line. Delete o
- A chat opens on its newest messages and brings in earlier ones as you scroll up, and no longer reloads after an app update.
- Tables, dividers, nested lists and inline code in replies are now formatted properly.
- Opening the app now shows an animated bot while it connects, so new messages are already in when the list appears.
+- On-device voice calls let you choose the language you speak (Voice chat settings → Language) instead of always following the system's first language.
diff --git a/docs/features/voice-call.md b/docs/features/voice-call.md
index 71fecd34..8bc05e9a 100644
--- a/docs/features/voice-call.md
+++ b/docs/features/voice-call.md
@@ -15,7 +15,9 @@ The bot is the same agent on the computer; the call only changes how you talk to
- Desktop: the capsule sits 16 CSS pixels below the window top and explicitly excludes itself
from Electron's window-drag region so the gear, mute and end buttons receive mouse clicks.
The iPhone places its capsule below native navigation; the terminal has no call controls.
-- Speech → text with `SFSpeechRecognizer` (on-device when the language supports it). A pause ends an
+- Speech → text with `SFSpeechRecognizer` (on-device when the language supports it) in the language
+ picked under Voice chat → Language (`callLanguage`; empty follows the system's first language,
+ which is often English on bilingual setups). A pause ends an
utterance and sends it as an ordinary message (`BotStore.send`), so it takes the normal path:
queued while the bot works, folded into its next turn. What you said appears in the chat as your
message; the bot's reply appears as its message.