Skip to content

Agent Zero AI framework

Fix TTS streaming race conditions and audio cancellation bugs

Frederic Thomasdd7ef855ce19
diff --git a/webui/components/chat/speech/speech-store.js b/webui/components/chat/speech/speech-store.js
index 74f4f115d..53aab8d51 100644
--- a/webui/components/chat/speech/speech-store.js
+++ b/webui/components/chat/speech/speech-store.js
@@ -205,6 +205,7 @@ const model = {
if (this.ttsStream.chunks.length == 0) return;
// if stream was already running, just updating chunks is enough
+ // The running loop will pick up the new chunks automatically
if (this.ttsStream.running) return;
else this.ttsStream.running = true; // proceed to running phase
@@ -214,22 +215,40 @@ const model = {
const spoken = [];
- // loop chunks from last spoken chunk index
- for (
- let i = this.ttsStream.lastChunkIndex + 1;
- i < this.ttsStream.chunks.length;
- i++
- ) {
+ // continuously loop until all chunks are spoken and stream is finished
+ while (true) {
+ // check if we should stop
+ if (terminator()) break;
+
+ // get the next chunk index to speak
+ const nextIndex = this.ttsStream.lastChunkIndex + 1;
+
+ // if no more chunks available, check if we should wait or exit
+ if (nextIndex >= this.ttsStream.chunks.length) {
+ // if stream is finished, we're done
+ if (this.ttsStream.finished) break;
+ // otherwise wait a bit for more chunks to arrive
+ await new Promise((resolve) => setTimeout(resolve, 50));
+ continue;
+ }
+
// do not speak the last chunk until finished (it is being generated)
- if (i == this.ttsStream.chunks.length - 1 && !this.ttsStream.finished)
- break;
+ if (
+ nextIndex == this.ttsStream.chunks.length - 1 &&
+ !this.ttsStream.finished
+ ) {
+ // wait a bit for more content or finish signal
+ await new Promise((resolve) => setTimeout(resolve, 50));
+ continue;
+ }
// set the index of last spoken chunk
- this.ttsStream.lastChunkIndex = i;
+ this.ttsStream.lastChunkIndex = nextIndex;
// speak the chunk
- spoken.push(this.ttsStream.chunks[i]);
- await this._speak(this.ttsStream.chunks[i], i > 0, () => terminator());
+ const chunk = this.ttsStream.chunks[nextIndex];
+ spoken.push(chunk);
+ await this._speak(chunk, nextIndex > 0, () => terminator());
}
// at the end, finish stream data
@@ -386,8 +405,8 @@ const model = {
while (waitForPrevious && this.isSpeaking) await sleep(25);
if (terminator && terminator()) return;
- // stop previous if any
- this.stopAudio();
+ // stop previous only if not waiting for it
+ if (!waitForPrevious) this.stopAudio();
this.browserUtterance = new SpeechSynthesisUtterance(text);
this.browserUtterance.onstart = () => {
@@ -410,8 +429,8 @@ const model = {
while (waitForPrevious && this.isSpeaking) await sleep(25);
if (terminator && terminator()) return;
- // stop previous if any
- this.stopAudio();
+ // stop previous only if not waiting for it
+ if (!waitForPrevious) this.stopAudio();
if (response.success) {
if (response.audio_parts) {
diff --git a/webui/components/messages/action-buttons/simple-action-buttons.js b/webui/components/messages/action-buttons/simple-action-buttons.js
index c19682d65..8f8d71222 100644
--- a/webui/components/messages/action-buttons/simple-action-buttons.js
+++ b/webui/components/messages/action-buttons/simple-action-buttons.js
@@ -97,7 +97,8 @@ export function addActionButtonsToElement(element) {
// Check if the button container is still fading in (opacity < 0.5)
if (parseFloat(window.getComputedStyle(container).opacity) < 0.5) return; // Don't proceed if still fading in
- const text = getTextContent(element);
+ // Get HTML content instead of plain text so cleanText() can properly handle code blocks
+ const text = getTextContent(element, true);
const icon = speakBtn.querySelector(".material-symbols-outlined");
if (!text || text.trim().length === 0) return;