Update app.py

app.py

@@ -557,131 +557,131 @@ second_of_silence = AudioSegment.silent() # use default
 second_of_silence.export("sil.wav", format='wav')
 
 
-# def generate_speech(history,chatbot_role):
-#     # Must set autoplay to True first
-#     yield (history, chatbot_role, "", wave_header_chunk() )
+def generate_speech(history,chatbot_role):
+    # Must set autoplay to True first
+    yield (history, chatbot_role, "", wave_header_chunk() )
 
-#     first_sentence=True 
-#     language="autodetect" # will predict from first sentence
+    first_sentence=True 
+    language="autodetect" # will predict from first sentence
     
-#     for sentence, history in get_sentence(history,chatbot_role):
-#         if sentence != "":
-#             if first_sentence:
-#                 language = detect_language(sentence) 
-#                 first_sentence=False 
+    for sentence, history in get_sentence(history,chatbot_role):
+        if sentence != "":
+            if first_sentence:
+                language = detect_language(sentence) 
+                first_sentence=False 
                 
-#             print("BG: inserting sentence to queue")
+            print("BG: inserting sentence to queue")
             
-#             generated_speech = generate_speech_for_sentence(history, chatbot_role, sentence,return_as_byte=True,language=language)
-#             if generated_speech is not None:
-#                 _, audio_dict = generated_speech
-#                 # We are using byte streaming
-#                 yield (history, chatbot_role, sentence, audio_dict["value"] )
+            generated_speech = generate_speech_for_sentence(history, chatbot_role, sentence,return_as_byte=True,language=language)
+            if generated_speech is not None:
+                _, audio_dict = generated_speech
+                # We are using byte streaming
+                yield (history, chatbot_role, sentence, audio_dict["value"] )
                 
             
-# # will generate speech audio file per sentence
-# def generate_speech_for_sentence(history, chatbot_role, sentence, return_as_byte=True, language="autodetect"):
+# will generate speech audio file per sentence
+def generate_speech_for_sentence(history, chatbot_role, sentence, return_as_byte=True, language="autodetect"):
    
-#     wav_bytestream = b""
+    wav_bytestream = b""
     
-#     if len(sentence)==0:
-#         print("EMPTY SENTENCE")
-#         return 
+    if len(sentence)==0:
+        print("EMPTY SENTENCE")
+        return 
     
-#     # Sometimes prompt </s> coming on output remove it
-#     # Some post process for speech only
-#     sentence = sentence.replace("</s>", "")
-#     # remove code from speech
-#     sentence = re.sub("```.*```", "", sentence, flags=re.DOTALL)
-#     sentence = re.sub("`.*`", "", sentence, flags=re.DOTALL)
+    # Sometimes prompt </s> coming on output remove it
+    # Some post process for speech only
+    sentence = sentence.replace("</s>", "")
+    # remove code from speech
+    sentence = re.sub("```.*```", "", sentence, flags=re.DOTALL)
+    sentence = re.sub("`.*`", "", sentence, flags=re.DOTALL)
     
-#     sentence = re.sub("\(.*\)", "", sentence, flags=re.DOTALL)
+    sentence = re.sub("\(.*\)", "", sentence, flags=re.DOTALL)
     
-#     sentence = sentence.replace("```", "")
-#     sentence = sentence.replace("...", " ")
-#     sentence = sentence.replace("(", " ")
-#     sentence = sentence.replace(")", " ")
-#     sentence = sentence.replace("<|assistant|>","")
-
-#     if len(sentence)==0:
-#         print("EMPTY SENTENCE after processing")
-#         return 
+    sentence = sentence.replace("```", "")
+    sentence = sentence.replace("...", " ")
+    sentence = sentence.replace("(", " ")
+    sentence = sentence.replace(")", " ")
+    sentence = sentence.replace("<|assistant|>","")
+
+    if len(sentence)==0:
+        print("EMPTY SENTENCE after processing")
+        return 
             
-#     # A fast fix for last chacter, may produce weird sounds if it is with text
-#     if (sentence[-1] in ["!", "?", ".", ","]) or (sentence[-2] in ["!", "?", ".", ","]):
-#         # just add a space
-#         sentence = sentence[:-1] + " " + sentence[-1]
-#     print("Sentence for speech:", sentence)
+    # A fast fix for last chacter, may produce weird sounds if it is with text
+    if (sentence[-1] in ["!", "?", ".", ","]) or (sentence[-2] in ["!", "?", ".", ","]):
+        # just add a space
+        sentence = sentence[:-1] + " " + sentence[-1]
+    print("Sentence for speech:", sentence)
 
     
-#     try:
-#         SENTENCE_SPLIT_LENGTH=350
-#         if len(sentence)<SENTENCE_SPLIT_LENGTH:
-#             # no problem continue on
-#             sentence_list = [sentence]
-#         else:
-#             # Until now nltk likely split sentences properly but we need additional 
-#             # check for longer sentence and split at last possible position
-#             # Do whatever necessary, first break at hypens then spaces and then even split very long words
-#             sentence_list=textwrap.wrap(sentence,SENTENCE_SPLIT_LENGTH)
-#             print("SPLITTED LONG SENTENCE:",sentence_list)
+    try:
+        SENTENCE_SPLIT_LENGTH=350
+        if len(sentence)<SENTENCE_SPLIT_LENGTH:
+            # no problem continue on
+            sentence_list = [sentence]
+        else:
+            # Until now nltk likely split sentences properly but we need additional 
+            # check for longer sentence and split at last possible position
+            # Do whatever necessary, first break at hypens then spaces and then even split very long words
+            sentence_list=textwrap.wrap(sentence,SENTENCE_SPLIT_LENGTH)
+            print("SPLITTED LONG SENTENCE:",sentence_list)
         
-#         for sentence in sentence_list:
+        for sentence in sentence_list:
             
-#             if any(c.isalnum() for c in sentence):
-#                 if language=="autodetect":
-#                     #on first call autodetect, nexts sentence calls will use same language
-#                     language = detect_language(sentence) 
+            if any(c.isalnum() for c in sentence):
+                if language=="autodetect":
+                    #on first call autodetect, nexts sentence calls will use same language
+                    language = detect_language(sentence) 
             
-#                 #exists at least 1 alphanumeric (utf-8) 
-#                 audio_stream = get_voice_streaming(
-#                         sentence, language, latent_map[chatbot_role]
-#                     )
-#             else:
-#                 # likely got a ' or " or some other text without alphanumeric in it
-#                 audio_stream = None 
+                #exists at least 1 alphanumeric (utf-8) 
+                audio_stream = get_voice_streaming(
+                        sentence, language, latent_map[chatbot_role]
+                    )
+            else:
+                # likely got a ' or " or some other text without alphanumeric in it
+                audio_stream = None 
                 
-#             # XTTS is actually using streaming response but we are playing audio by sentence
-#             # If you want direct XTTS voice streaming (send each chunk to voice ) you may set DIRECT_STREAM=1 environment variable
-#             if audio_stream is not None:
-#                 wav_chunks = wave_header_chunk()
-#                 frame_length = 0
-#                 for chunk in audio_stream:
-#                     try:
-#                         wav_bytestream += chunk
-#                         wav_chunks += chunk
-#                         frame_length += len(chunk)
-#                     except:
-#                         # hack to continue on playing. sometimes last chunk is empty , will be fixed on next TTS
-#                         continue
-
-#             if audio_stream is not None:
-#                 if not return_as_byte:
-#                     audio_unique_filename = "/tmp/"+ str(uuid.uuid4())+".wav"
-#                     with open(audio_unique_filename, "wb") as f:
-#                         f.write(wav_chunks)
-#                     #Will write filename to context variable
-#                     return (history , gr.Audio.update(value=audio_unique_filename, autoplay=True))
-#                 else:
-#                     return (history , gr.Audio.update(value=wav_bytestream, autoplay=True))
-#     except RuntimeError as e:
-#         if "device-side assert" in str(e):
-#             # cannot do anything on cuda device side error, need tor estart
-#             print(
-#                 f"Exit due to: Unrecoverable exception caused by prompt:{sentence}",
-#                 flush=True,
-#             )
-#             gr.Warning("Unhandled Exception encounter, please retry in a minute")
-#             print("Cuda device-assert Runtime encountered need restart")
-
-#             # HF Space specific.. This error is unrecoverable need to restart space
-#             api.restart_space(repo_id=repo_id)
-#         else:
-#             print("RuntimeError: non device-side assert error:", str(e))
-#             raise e
-
-#     print("All speech ended")
-#     return 
+            # XTTS is actually using streaming response but we are playing audio by sentence
+            # If you want direct XTTS voice streaming (send each chunk to voice ) you may set DIRECT_STREAM=1 environment variable
+            if audio_stream is not None:
+                wav_chunks = wave_header_chunk()
+                frame_length = 0
+                for chunk in audio_stream:
+                    try:
+                        wav_bytestream += chunk
+                        wav_chunks += chunk
+                        frame_length += len(chunk)
+                    except:
+                        # hack to continue on playing. sometimes last chunk is empty , will be fixed on next TTS
+                        continue
+
+            if audio_stream is not None:
+                if not return_as_byte:
+                    audio_unique_filename = "/tmp/"+ str(uuid.uuid4())+".wav"
+                    with open(audio_unique_filename, "wb") as f:
+                        f.write(wav_chunks)
+                    #Will write filename to context variable
+                    return (history , gr.Audio.update(value=audio_unique_filename, autoplay=True))
+                else:
+                    return (history , gr.Audio.update(value=wav_bytestream, autoplay=True))
+    except RuntimeError as e:
+        if "device-side assert" in str(e):
+            # cannot do anything on cuda device side error, need tor estart
+            print(
+                f"Exit due to: Unrecoverable exception caused by prompt:{sentence}",
+                flush=True,
+            )
+            gr.Warning("Unhandled Exception encounter, please retry in a minute")
+            print("Cuda device-assert Runtime encountered need restart")
+
+            # HF Space specific.. This error is unrecoverable need to restart space
+            api.restart_space(repo_id=repo_id)
+        else:
+            print("RuntimeError: non device-side assert error:", str(e))
+            raise e
+
+    print("All speech ended")
+    return 
 
 
 latent_map = {}