Merge pull request #3714 from rudycelekli/fix/cow50-voice-identical-text-filenames-20261012

fix(voice): isolate output files for repeated identical speech

tests/test_voice_identical_text_files.py

@@ -0,0 +1,40 @@
+import importlib
+import importlib.util
+from pathlib import Path
+import sys
+import time
+from types import SimpleNamespace
+import pytest
+from bridge.reply import ReplyType
+
+@pytest.mark.parametrize('provider', ['google', 'edge', 'pytts'])
+def test_identical_replies_have_independent_output_files(tmp_path, monkeypatch, provider):
+    if provider in ('google', 'pytts'):
+        filename = 'test_google_voice_errors.py' if provider == 'google' else 'test_pytts_voice_timeout.py'
+        spec = importlib.util.spec_from_file_location('voice_' + provider + '_support', Path(__file__).with_name(filename))
+        support = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(support)
+    else:
+        monkeypatch.setitem(sys.modules, 'edge_tts', SimpleNamespace())
+    module = importlib.import_module('voice.' + provider + '.' + provider + '_voice')
+    monkeypatch.setattr(time, 'time', lambda: 1000)
+    monkeypatch.setattr(module.TmpDir, 'path', lambda self: str(tmp_path) + '/')
+    if provider == 'google':
+        monkeypatch.setattr(module, 'gTTS', lambda **kw: SimpleNamespace(save=lambda path: Path(path).write_bytes(b'audio')))
+        voice = module.GoogleVoice()
+    elif provider == 'edge':
+        voice = module.EdgeVoice()
+        async def generate(text, path):
+            Path(path).write_bytes(b'audio')
+        monkeypatch.setattr(voice, 'gen_voice', generate)
+    else:
+        voice = module.PyttsVoice()
+        engine = support._FakeEngine()
+        engine.save_to_file = lambda text, path: Path(path).write_bytes(b'audio')
+        voice.engine = engine
+    first = voice.textToVoice('same reply')
+    Path(first.content).write_bytes(b'first-clip-sentinel')
+    second = voice.textToVoice('same reply')
+    assert first.type == second.type == ReplyType.VOICE
+    assert first.content != second.content
+    assert Path(first.content).read_bytes() == b'first-clip-sentinel'

voice/ali/ali_api.py

@@ -10,7 +10,6 @@ Description:
 
 import http.client
 import json
-import time
 import requests
 import datetime
 import hashlib
@@ -51,7 +50,7 @@ def text_to_speech_aliyun(url, text, appkey, token):
     response = requests.post(url, headers=headers, data=json.dumps(data), timeout=(5, 60), stream=True)
 
     if response.status_code == 200 and response.headers['Content-Type'] == 'audio/mpeg':
-        output_file = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".wav"
+        output_file = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".wav"
 
         try:
             save_response(response, output_file, MAX_FILE_BYTES)

voice/azure/azure_voice.py

@@ -3,7 +3,7 @@ azure voice service
 """
 import json
 import os
-import time
+import uuid
 
 import azure.cognitiveservices.speech as speechsdk
 from langid import classify
@@ -81,7 +81,7 @@ class AzureVoice(Voice):
         else:
             self.speech_config.speech_synthesis_voice_name = self.config["speech_synthesis_voice_name"]
         # Avoid the same filename under multithreading
-        fileName = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".wav"
+        fileName = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".wav"
         audio_config = speechsdk.AudioConfig(filename=fileName)
         speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=self.speech_config, audio_config=audio_config)
         result = speech_synthesizer.speak_text(text)

voice/baidu/baidu_voice.py

@@ -6,6 +6,7 @@ import os
 import time
 import threading
 import requests
+import uuid
 
 from aip import AipSpeech
 
@@ -157,7 +158,7 @@ class BaiduVoice(Voice):
         # used to be buffered whole by `.content` and written out before
         # anything could object. `download_to_file` also writes through a temp
         # file, so an oversized or truncated download leaves nothing behind.
-        fn = TmpDir().path() + f"reply-long-{int(time.time())}-{hash(text)&0x7FFFFFFF}.mp3"
+        fn = TmpDir().path() + f"reply-long-{uuid.uuid4().hex}.mp3"
         try:
             download_to_file(audio_url, fn, MAX_FILE_BYTES, timeout=REQUEST_TIMEOUT)
         except Exception as e:
@@ -177,7 +178,7 @@ class BaiduVoice(Voice):
                     {"spd":self.spd, "pit":self.pit, "vol":self.vol, "per":self.per}
                 )
                 if not isinstance(result, dict):
-                    fn = TmpDir().path() + f"reply-{int(time.time())}-{hash(text)&0x7FFFFFFF}.mp3"
+                    fn = TmpDir().path() + f"reply-{uuid.uuid4().hex}.mp3"
                     with open(fn, "wb") as f:
                         f.write(result)
                     logger.info("[Baidu] 短文本合成 success: %s", fn)

voice/edge/edge_voice.py

@@ -1,5 +1,5 @@
 # Requires: edge-tts  (pip install edge-tts)
-import time
+import uuid
 
 import edge_tts
 import asyncio
@@ -43,7 +43,7 @@ class EdgeVoice(Voice):
         await communicate.save(fileName)
 
     def textToVoice(self, text):
-        fileName = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".mp3"
+        fileName = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".mp3"
 
         try:
             asyncio.run(self.gen_voice(text, fileName))

voice/elevent/elevent_voice.py

@@ -1,5 +1,5 @@
 # Requires: elevenlabs==1.0.3  (pip install elevenlabs==1.0.3)
-import time
+import uuid
 
 from elevenlabs.client import ElevenLabs
 from elevenlabs import save
@@ -22,7 +22,7 @@ class ElevenLabsVoice(Voice):
         pass
 
     def textToVoice(self, text):
-        fileName = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".mp3"
+        fileName = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".mp3"
 
         try:
             audio = client.generate(

voice/google/google_voice.py

@@ -2,7 +2,7 @@
 google voice service
 """
 
-import time
+import uuid
 
 import speech_recognition
 from gtts import gTTS
@@ -39,7 +39,7 @@ class GoogleVoice(Voice):
     def textToVoice(self, text):
         try:
             # Avoid the same filename under multithreading
-            mp3File = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".mp3"
+            mp3File = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".mp3"
             tts = gTTS(text=text, lang="zh")
             tts.save(mp3File)
             logger.info("[Google] textToVoice text={} voice file name={}".format(text, mp3File))

voice/pytts/pytts_voice.py

@@ -5,6 +5,7 @@ pytts voice service (offline)
 import os
 import sys
 import time
+import uuid
 
 import pyttsx3
 
@@ -41,7 +42,7 @@ class PyttsVoice(Voice):
     def textToVoice(self, text):
         try:
             # Avoid the same filename under multithreading
-            wavFileName = "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".wav"
+            wavFileName = "reply-" + uuid.uuid4().hex + ".wav"
             wavFile = TmpDir().path() + wavFileName
             logger.info("[Pytts] textToVoice text={} voice file name={}".format(text, wavFile))
 

voice/tencent/tencent_voice.py

@@ -2,6 +2,7 @@ import json
 import base64
 import os
 import time
+import uuid
 from voice.voice import Voice
 from common.log import logger
 from tencentcloud.common import credential
@@ -105,7 +106,7 @@ class TencentVoice(Voice):
             response = client.TextToVoice(req)
             
             if response.Audio:
-                fileName = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".mp3"
+                fileName = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".mp3"
                 with open(fileName, "wb") as f:
                     f.write(base64.b64decode(response.Audio))
                 logger.info("[Tencent] textToVoice text={} voice file name={}".format(text, fileName))

voice/xunfei/xunfei_voice.py

@@ -29,6 +29,7 @@ from voice.voice import Voice
 from .xunfei_asr import xunfei_asr
 from .xunfei_tts import xunfei_tts
 import shutil
+import uuid
 
 try:
     from voice.audio_convert import any_to_mp3
@@ -81,7 +82,7 @@ class XunfeiVoice(Voice):
     def textToVoice(self, text):
         try:
             # Avoid the same filename under multithreading
-            fileName = TmpDir().path() + "reply-" + str(int(time.time())) + "-" + str(hash(text) & 0x7FFFFFFF) + ".mp3"
+            fileName = TmpDir().path() + "reply-" + uuid.uuid4().hex + ".mp3"
             xunfei_tts(self.APPID,self.APIKey,self.APISecret,self.BusinessArgsTTS,text,fileName)
             logger.info("[Xunfei] textToVoice text={} voice file name={}".format(text, fileName))
             reply = Reply(ReplyType.VOICE, fileName)