Update app.py

app.py

@@ -24,6 +24,7 @@ from googlemaps import Client as GoogleMapsClient
 from gtts import gTTS
 from diffusers import StableDiffusion3Pipeline
 import soundfile as sf
+import numpy as np
 
 from langchain_openai import OpenAIEmbeddings, ChatOpenAI
 from langchain_pinecone import PineconeVectorStore
@@ -283,20 +284,74 @@ def bot(history, choice, tts_model):
 
     # Generate audio and process output prompt in parallel
     with concurrent.futures.ThreadPoolExecutor() as executor:
-        audio_future = executor.submit(generate_audio, tts_model, response)
-        for character in response:
-            history[-1][1] += character
-            time.sleep(0.05)  # Adjust the speed of text appearance
-            yield history, None
+        audio_future = executor.submit(generate_audio_chunks, tts_model, response)
+        text_chunks = [response[i:i + 100] for i in range(0, len(response), 100)]  # Chunk the text for streaming
+        for text_chunk in text_chunks:
+            history[-1][1] += text_chunk
+            audio_chunk = next(audio_future.result(), None)
+            time.sleep(0.2)  # Adjust this to synchronize text and audio appearance
+            yield history, audio_chunk
         
-        audio_path = audio_future.result()
-        yield history, audio_path
+        for remaining_audio_chunk in audio_future.result():
+            yield history, remaining_audio_chunk
 
-def generate_audio(tts_model, text):
+def generate_audio_chunks(tts_model, text):
     if tts_model == "ElevenLabs":
-        return generate_audio_elevenlabs(text)
+        return generate_audio_elevenlabs_chunks(text)
+    else:
+        return generate_audio_parler_tts_chunks(text)
+
+def generate_audio_elevenlabs_chunks(text):
+    XI_API_KEY = os.environ['ELEVENLABS_API']
+    VOICE_ID = 'd9MIrwLnvDeH7aZb61E9'  # Replace with your voice ID
+    tts_url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream"
+    headers = {
+        "Accept": "application/json",
+        "xi-api-key": XI_API_KEY
+    }
+    data = {
+        "text": str(text),
+        "model_id": "eleven_multilingual_v2",
+        "voice_settings": {
+            "stability": 1.0,
+            "similarity_boost": 0.0,
+            "style": 0.60,  # Adjust style for more romantic tone
+            "use_speaker_boost": False
+        }
+    }
+    response = requests.post(tts_url, headers=headers, json=data, stream=True)
+    if response.ok:
+        for chunk in response.iter_content(chunk_size=1024):
+            yield chunk
     else:
-        return generate_audio_parler_tts(text)
+        logging.error(f"Error generating audio: {response.text}")
+        return None
+
+def generate_audio_parler_tts_chunks(text):
+    model_id = 'parler-tts/parler_tts_mini_v0.1'
+    device = "cuda:0" if torch.cuda.is_available() else "cpu"
+    try:
+        model = ParlerTTSForConditionalGeneration.from_pretrained(model_id).to(device)
+    except torch.cuda.OutOfMemoryError:
+        print("CUDA out of memory. Switching to CPU.")
+        device = "cpu"
+        model = ParlerTTSForConditionalGeneration.from_pretrained(model_id).to(device)
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+    description = "A female speaker with a slightly low-pitched voice delivers her words quite expressively, in a very confined sounding environment with clear audio quality. She speaks very fast."
+
+    input_ids = tokenizer(description, return_tensors="pt").input_ids.to(device)
+    prompt_input_ids = tokenizer(text, return_tensors="pt").input_ids.to(device)
+
+    generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids, max_new_tokens=200)
+    audio_arr = generation.cpu().numpy().squeeze()
+
+    chunk_size = 16000  # Define a chunk size (adjust as needed)
+    for i in range(0, len(audio_arr), chunk_size):
+        audio_chunk = audio_arr[i:i + chunk_size]
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
+            sf.write(f.name, audio_chunk, model.config.sampling_rate)
+            yield f.name
 
 def add_message(history, message):
     history.append((message, None))
@@ -350,7 +405,7 @@ def fetch_local_news():
     api_key = os.environ['SERP_API']
     url = f'https://serpapi.com/search.json?engine=google_news&q=birmingham headline&api_key={api_key}'
     response = requests.get(url)
-    if response.status_code == 200:
+    if response.status_code == 200):
         results = response.json().get("news_results", [])
         news_html = """
         <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Birmingham Today</h2>
@@ -478,62 +533,6 @@ def show_map_if_details(history,choice):
     else:
         return gr.update(visible=False), ""
 
-def generate_audio_elevenlabs(text):
-    XI_API_KEY = os.environ['ELEVENLABS_API']
-    VOICE_ID = 'd9MIrwLnvDeH7aZb61E9'  # Replace with your voice ID
-    tts_url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream"
-    headers = {
-        "Accept": "application/json",
-        "xi-api-key": XI_API_KEY
-    }
-    data = {
-        "text": str(text),
-        "model_id": "eleven_multilingual_v2",
-        "voice_settings": {
-            "stability": 1.0,
-            "similarity_boost": 0.0,
-            "style": 0.60,  # Adjust style for more romantic tone
-            "use_speaker_boost": False
-        }
-    }
-    response = requests.post(tts_url, headers=headers, json=data, stream=True)
-    if response.ok:
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f:
-            for chunk in response.iter_content(chunk_size=1024):
-                f.write(chunk)
-            temp_audio_path = f.name
-        logging.debug(f"Audio saved to {temp_audio_path}")
-        return temp_audio_path
-    else:
-        logging.error(f"Error generating audio: {response.text}")
-        return None
-
-def generate_audio_parler_tts(text):
-    model_id = 'parler-tts/parler_tts_mini_v0.1'
-    device = "cuda:0" if torch.cuda.is_available() else "cpu"
-    try:
-        model = ParlerTTSForConditionalGeneration.from_pretrained(model_id).to(device)
-    except torch.cuda.OutOfMemoryError:
-        print("CUDA out of memory. Switching to CPU.")
-        device = "cpu"
-        model = ParlerTTSForConditionalGeneration.from_pretrained(model_id).to(device)
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-    description = "A female speaker with a slightly low-pitched voice delivers her words quite expressively, in a very confined sounding environment with clear audio quality. She speaks very fast."
-
-    input_ids = tokenizer(description, return_tensors="pt").input_ids.to(device)
-    prompt_input_ids = tokenizer(text, return_tensors="pt").input_ids.to(device)
-
-    generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids, max_new_tokens=200)
-    audio_arr = generation.cpu().numpy().squeeze()
-    
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
-        sf.write(f.name, audio_arr, model.config.sampling_rate)
-        temp_audio_path = f.name
-
-    logging.debug(f"Audio saved to {temp_audio_path}")
-    return temp_audio_path
-
 # Stable Diffusion setup
 pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
 pipe = pipe.to("cuda")