Add MP3 output option for TTS

app.py

@@ -3,6 +3,7 @@ import io
 import json
 import os
 import shutil
+import subprocess
 import sys
 import tempfile
 import threading
@@ -77,10 +78,45 @@ def decode_audio_b64(audio_base64: str, filename: str) -> Path:
         raise gr.Error("invalid base64 audio")
 
 
-def wav_bytes_to_b64(audio, sample_rate: int = 24000) -> str:
-    buf = io.BytesIO()
-    sf.write(buf, audio, sample_rate, format="WAV")
-    return base64.b64encode(buf.getvalue()).decode("ascii")
+def audio_bytes_to_b64(audio, sample_rate: int = 24000, audio_format: str = "wav") -> tuple[str, str]:
+    audio_format = (audio_format or "wav").lower().strip()
+    if audio_format not in {"wav", "mp3"}:
+        raise gr.Error("audio_format must be wav or mp3")
+
+    wav_buf = io.BytesIO()
+    sf.write(wav_buf, audio, sample_rate, format="WAV")
+    wav_bytes = wav_buf.getvalue()
+    if audio_format == "wav":
+        return base64.b64encode(wav_bytes).decode("ascii"), "audio/wav"
+
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as wav_file:
+        wav_file.write(wav_bytes)
+        wav_path = Path(wav_file.name)
+    mp3_path = wav_path.with_suffix(".mp3")
+    try:
+        subprocess.run(
+            [
+                "ffmpeg",
+                "-y",
+                "-hide_banner",
+                "-loglevel",
+                "error",
+                "-i",
+                str(wav_path),
+                "-codec:a",
+                "libmp3lame",
+                "-b:a",
+                "128k",
+                str(mp3_path),
+            ],
+            check=True,
+        )
+        return base64.b64encode(mp3_path.read_bytes()).decode("ascii"), "audio/mpeg"
+    except Exception as exc:
+        raise gr.Error(f"mp3 encoding failed: {exc}") from exc
+    finally:
+        wav_path.unlink(missing_ok=True)
+        mp3_path.unlink(missing_ok=True)
 
 
 def get_asr_pipeline():
@@ -177,7 +213,13 @@ def transcribe_b64(
     )
 
 
-def speech_b64(api_key: str, text: str, voice: str = "zf_xiaoxiao", speed: float = 1.0) -> str:
+def speech_b64(
+    api_key: str,
+    text: str,
+    voice: str = "zf_xiaoxiao",
+    speed: float = 1.0,
+    audio_format: str = "wav",
+) -> str:
     check_key(api_key)
     speed = float(speed)
     if not text.strip():
@@ -197,12 +239,14 @@ def speech_b64(api_key: str, text: str, voice: str = "zf_xiaoxiao", speed: float
     import numpy as np
 
     audio = np.concatenate(chunks)
+    audio_base64, mime_type = audio_bytes_to_b64(audio, 24000, audio_format)
     return json.dumps(
         {
-            "audio_base64": wav_bytes_to_b64(audio, 24000),
-            "mime_type": "audio/wav",
+            "audio_base64": audio_base64,
+            "mime_type": mime_type,
             "sample_rate": 24000,
             "voice": voice,
+            "audio_format": audio_format,
         },
         ensure_ascii=False,
     )
@@ -238,10 +282,11 @@ with gr.Blocks(title="Voice API") as demo:
         text = gr.Textbox(label="text", lines=4)
         voice = gr.Textbox(label="voice", value="zf_xiaoxiao")
         speed = gr.Number(label="speed", value=1.0)
+        audio_format = gr.Textbox(label="audio_format", value="wav")
         tts_out = gr.Textbox(label="result", lines=8)
         gr.Button("Synthesize").click(
             speech_b64,
-            inputs=[api_key, text, voice, speed],
+            inputs=[api_key, text, voice, speed, audio_format],
             outputs=[tts_out],
             api_name="speech_b64",
         )