Update app.py

app.py

@@ -6,7 +6,7 @@ import numpy as np
 import wave
 
 #tts
-from balacoon_tts import TTS, SpeechUtterance
+from balacoon_tts import TTS
 from threading import Lock
 from huggingface_hub import hf_hub_download, list_repo_files
 from pydub import AudioSegment
@@ -28,9 +28,6 @@ for name in list_repo_files(repo_id="balacoon/tts"):
                 local_dir=os.getcwd(),
             )
 
-# locker that disallow access to the tts object from more then one thread
-locker = Lock()
-
 #client
 client = Groq(
     api_key=os.getenv("GROQ_API_KEY"),
@@ -101,15 +98,14 @@ async def greet(product,description):
     if response.choices[0].message.content != "not moderated":
         a_list = ["Sorry, I can't proceed for generating marketing email. Your content needs to be moderated first. Thank you!"]
         tts = TTS(os.path.join(os.getcwd(), tts_model_str))
-        # create an utterance, which will hold the syntheis state
-        utterance = SpeechUtterance(a_list[0])
-        # run synthesis in a loop
-        while True:
-            samples = tts.synthesize_chunk(utterance, tts.get_speakers()[-1])
-            if len(samples) == 0:
-                # all the samples were already generated
-                break
-            yield gr.Audio(value=(tts.get_sampling_rate(), samples))
+        speakers = tts.get_speakers()
+        if len(a_list[0]) > 1024:
+            # truncate the text
+            text_str = a_list[0][:1024]
+        else:
+            text_str = a_list[0]
+        samples = tts.synthesize(text_str, speakers[-1])
+        yield gr.Audio(value=(tts.get_sampling_rate(), samples))
     else:
         output = llm.create_chat_completion(
         messages=[
@@ -128,16 +124,11 @@ async def greet(product,description):
         for chunk in output:
             delta = chunk['choices'][0]['delta']
             if 'content' in delta:
-                partial_message = partial_message + delta.get('content', '')
-        # create an utterance, which will hold the syntheis state
-        utterance = SpeechUtterance(partial_message)
-        # run synthesis in a loop
-        while True:
-            samples = tts.synthesize_chunk(utterance, tts.get_speakers()[-1])
-            if len(samples) == 0:
-                # all the samples were already generated
-                break
-            yield gr.Audio(value=(tts.get_sampling_rate(), samples))
+                tts = TTS(os.path.join(os.getcwd(), tts_model_str))
+                speakers = tts.get_speakers()
+                samples = tts.synthesize(delta.get('content', ''), speakers[-1])
+        yield (tts.get_sampling_rate(), samples)
+
 
 demo = gr.Interface(fn=greet, inputs=["text","text"], outputs=gr.Audio(), concurrency_limit=10)
 demo.launch()