Merge pull request #11 from teddybear082/fix_opus_output_support

README.md

@@ -154,7 +154,7 @@ with client.audio.speech.with_streaming_response.create(
 | `model`           | string  | No       | -       | Ignored (for OpenAI compatibility)                 |
 | `input`           | string  | Yes      | -       | Text to synthesize                                 |
 | `voice`           | string  | No       | `alba`  | Voice ID (see `/v1/voices`)                        |
-| `response_format` | string  | No       | `mp3`   | Output format: `mp3`, `wav`, `pcm`, `opus`, `flac` |
+| `response_format` | string  | No       | `mp3`   | Output format: `mp3`, `wav`, `pcm`, `opus`, `aac`, `flac` |
 | `stream`          | boolean | No       | `false` | Enable streaming response                          |
 
 ## Custom Voices

app/services/audio.py

@@ -63,12 +63,34 @@ def convert_audio(
     if audio_tensor.dim() == 1:
         audio_tensor = audio_tensor.unsqueeze(0)
 
+    # Handle PCM raw bytes (no container)
+    if target_format == 'pcm':
+        try:
+            pcm_bytes = tensor_to_pcm_bytes(audio_tensor)
+            buffer.write(pcm_bytes)
+            buffer.seek(0)
+            return buffer
+        except Exception as e:
+            logger.error(f'Error converting audio to PCM: {e}')
+            raise
+
+    # Map OpenAI format names to torchaudio/backend supported format names
+    # torchaudio uses 'ogg' as the container for 'opus'
+    # 'aac' usually requires 'adts' or 'm4a'
+    actual_format = target_format
+    if actual_format == 'opus':
+        actual_format = 'ogg'
+    elif actual_format == 'aac':
+        actual_format = 'adts'
+
     try:
-        torchaudio.save(buffer, audio_tensor, sample_rate, format=target_format)
+        torchaudio.save(buffer, audio_tensor, sample_rate, format=actual_format)
         buffer.seek(0)
         return buffer
     except Exception as e:
-        logger.error(f'Error converting audio to {target_format}: {e}')
+        logger.error(
+            f'Error converting audio to {target_format} (backend format: {actual_format}): {e}'
+        )
         raise
 
 

pyproject.toml

@@ -4,7 +4,7 @@ license = {text = "MIT"}
 name = "pocket-tts-openai-server"
 readme = "README.md"
 requires-python = ">=3.10"
-version = "2.5.1"
+version = "2.5.2"
 
 dependencies = [
   "flask>=3.0.0",
@@ -12,6 +12,7 @@ dependencies = [
   "pocket-tts>=2.0.0",
   "torch>=2.0.0,<=2.8.0",
   "torchaudio>=2.0.0,<=2.8.0",
+  "torchao==0.13.0",
   "scipy>=1.10.0",
   "numpy>=1.24.0",
   "soundfile>=0.12.0",

requirements.txt

@@ -9,6 +9,7 @@ pocket-tts>=2.0.0
 # Audio processing
 torch>=2.0.0,<=2.8.0
 torchaudio>=2.0.0,<=2.8.0
+torchao==0.13.0
 scipy>=1.10.0
 numpy>=1.24.0
 soundfile>=0.12.0

static/js/app.js

@@ -43,6 +43,8 @@ document.addEventListener('DOMContentLoaded', async () => {
 			if (fmt === 'mp3') {
 				infoLabel.textContent =
 					'Streaming is not available for MP3 (Server limitation). A full file will be generated and played.';
+			} else if (['opus', 'aac', 'flac'].includes(fmt)) {
+				infoLabel.textContent = `Streaming is not available for ${fmt.toUpperCase()}. A full file will be generated and played.`;
 			} else {
 				infoLabel.textContent = 'Streaming is not available for this format.';
 			}

templates/index.html

@@ -126,6 +126,9 @@ Hello! I am Pocket TTS, running locally on your machine. I can clone voices and
 						"
 					>
 						<option value="mp3">MP3</option>
+						<option value="opus">Opus</option>
+						<option value="aac">AAC</option>
+						<option value="flac">FLAC</option>
 						<option value="wav" selected>WAV</option>
 						<option value="pcm">PCM (Raw)</option>
 					</select>
@@ -272,8 +275,9 @@ Hello! I am Pocket TTS, running locally on your machine. I can clone voices and
 									<td></td>
 									<td><code>mp3</code></td>
 									<td>
-										Audio format: <code>mp3</code>, <code>wav</code>,
-										<code>pcm</code>, <code>opus</code>, <code>flac</code>
+										Audio format: <code>mp3</code>, <code>opus</code>,
+										<code>aac</code>, <code>flac</code>, <code>wav</code>,
+										<code>pcm</code>
 									</td>
 								</tr>
 								<tr>