Added Temp, Top k, Top p

app.py

@@ -43,27 +43,42 @@ examples = [
     [
         "मुले बागेत खेळत आहेत आणि पक्षी किलबिलाट करत आहेत.",
         "Sunita speaks slowly in a calm, moderate-pitched voice, delivering the news with a neutral tone. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "ಉದ್ಯಾನದಲ್ಲಿ ಮಕ್ಕಳ ಆಟವಾಡುತ್ತಿದ್ದಾರೆ ಮತ್ತು ಪಕ್ಷಿಗಳು ಚಿಲಿಪಿಲಿ ಮಾಡುತ್ತಿವೆ.",
         "Suresh speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "বাচ্চারা বাগানে খেলছে আর পাখি কিচিরমিচির করছে।",
         "Aditi speaks at a moderate pace and pitch, with a clear, neutral tone and no emotional emphasis. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "పిల్లలు తోటలో ఆడుకుంటున్నారు, పక్షుల కిలకిలరావాలు.",
         "Prakash speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "పిల్లలు తోటలో ఆడుకుంటున్నారు, పక్షుల కిలకిలరావాలు.",
         "Prakash speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "This is the best time of my life, Bartley,' she said happily",
@@ -73,22 +88,34 @@ examples = [
     [
         "Montrose also, after having experienced still more variety of good and bad fortune, threw down his arms, and retired out of the kingdom.",
         "A female speaker with a slightly low-pitched, quite monotone voice speaks with an American accent at a slightly faster-than-average pace in a confined space with very clear audio.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "बगीचे में बच्चे खेल रहे हैं और पक्षी चहचहा रहे हैं।",
         "Rohit speaks with a slightly high-pitched voice delivering his words at a slightly slow pace in a small, confined space with a touch of background noise and a quite monotone tone.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "കുട്ടികൾ പൂന്തോട്ടത്തിൽ കളിക്കുന്നു, പക്ഷികൾ ചിലയ്ക്കുന്നു.",
         "Anjali speaks with a low-pitched voice delivering her words at a fast pace and an animated tone, in a very spacious environment, accompanied by noticeable background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "குழந்தைகள் தோட்டத்தில் விளையாடுகிறார்கள், பறவைகள் கிண்டல் செய்கின்றன.",
         "Jaya speaks with a slightly low-pitched, quite monotone voice at a slightly faster-than-average pace in a confined space with very clear audio.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ]
 ]
 
@@ -97,64 +124,91 @@ finetuned_examples = [
     [
         "मुले बागेत खेळत आहेत आणि पक्षी किलबिलाट करत आहेत.",
         "Sunita speaks slowly in a calm, moderate-pitched voice, delivering the news with a neutral tone. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "ಉದ್ಯಾನದಲ್ಲಿ ಮಕ್ಕಳ ಆಟವಾಡುತ್ತಿದ್ದಾರೆ ಮತ್ತು ಪಕ್ಷಿಗಳು ಚಿಲಿಪಿಲಿ ಮಾಡುತ್ತಿವೆ.",
         "Suresh speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "বাচ্চারা বাগানে খেলছে আর পাখি কিচিরমিচির করছে।",
         "Aditi speaks at a moderate pace and pitch, with a clear, neutral tone and no emotional emphasis. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "పిల్లలు తోటలో ఆడుకుంటున్నారు, పక్షుల కిలకిలరావాలు.",
         "Prakash speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "పిల్లలు తోటలో ఆడుకుంటున్నారు, పక్షుల కిలకిలరావాలు.",
         "Prakash speaks slowly in a low-pitched, calm voice, with a neutral tone, perfect for narration. The recording is very high quality with no background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "This is the best time of my life, Bartley,' she said happily",
         "A male speaker with a low-pitched voice speaks with a British accent at a fast pace in a small, confined space with very clear audio and an animated tone.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "Montrose also, after having experienced still more variety of good and bad fortune, threw down his arms, and retired out of the kingdom.",
         "A female speaker with a slightly low-pitched, quite monotone voice speaks with an American accent at a slightly faster-than-average pace in a confined space with very clear audio.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "बगीचे में बच्चे खेल रहे हैं और पक्षी चहचहा रहे हैं।",
         "Rohit speaks with a slightly high-pitched voice delivering his words at a slightly slow pace in a small, confined space with a touch of background noise and a quite monotone tone.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "കുട്ടികൾ പൂന്തോട്ടത്തിൽ കളിക്കുന്നു, പക്ഷികൾ ചിലയ്ക്കുന്നു.",
         "Anjali speaks with a low-pitched voice delivering her words at a fast pace and an animated tone, in a very spacious environment, accompanied by noticeable background noise.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ],
     [
         "குழந்தைகள் தோட்டத்தில் விளையாடுகிறார்கள், பறவைகள் கிண்டல் செய்கின்றன.",
         "Jaya speaks with a slightly low-pitched, quite monotone voice at a slightly faster-than-average pace in a confined space with very clear audio.",
-        3.0
+        3.0,
+        0.8,
+        0.9,
+        50
     ]
 ]
 
-
 def numpy_to_mp3(audio_array, sampling_rate):
-    # Normalize audio_array if it's floating-point
     if np.issubdtype(audio_array.dtype, np.floating):
         max_val = np.max(np.abs(audio_array))
-        audio_array = (audio_array / max_val) * 32767  # Normalize to 16-bit range
+        audio_array = (audio_array / max_val) * 32767
         audio_array = audio_array.astype(np.int16)
 
-    # Create an audio segment from the numpy array
     audio_segment = AudioSegment(
         audio_array.tobytes(),
         frame_rate=sampling_rate,
@@ -162,11 +216,9 @@ def numpy_to_mp3(audio_array, sampling_rate):
         channels=1
     )
 
-    # Export the audio segment to MP3 bytes - use a high bitrate to maximise quality
     mp3_io = io.BytesIO()
     audio_segment.export(mp3_io, format="mp3", bitrate="320k")
 
-    # Get the MP3 bytes
     mp3_bytes = mp3_io.getvalue()
     mp3_io.close()
 
@@ -176,14 +228,12 @@ sampling_rate = model.audio_encoder.config.sampling_rate
 frame_rate = model.audio_encoder.config.frame_rate
 
 @spaces.GPU
-def generate_base(text, description,):
-    # Initialize variables
-    chunk_size = 25  # Process max 25 words or a sentence at a time
+def generate_base(text, description, temperature, top_p, top_k):
+    chunk_size = 25
     
-    # Tokenize the full text and description
     inputs = description_tokenizer(description, return_tensors="pt").to(device)
 
-    sentences_text = nltk.sent_tokenize(text) # this gives us a list of sentences
+    sentences_text = nltk.sent_tokenize(text)
     curr_sentence = ""
     chunks = []
     for sentence in sentences_text:
@@ -201,22 +251,21 @@ def generate_base(text, description,):
 
     all_audio = []
     
-    # Process each chunk
     for chunk in chunks:
-        # Tokenize the chunk
         prompt = tokenizer(chunk, return_tensors="pt").to(device)
         
-        # Generate audio for the chunk
         generation = model.generate(
             input_ids=inputs.input_ids,
             attention_mask=inputs.attention_mask,
             prompt_input_ids=prompt.input_ids,
             prompt_attention_mask=prompt.attention_mask,
             do_sample=True,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
             return_dict_in_generate=True
         )
             
-        # Extract audio from generation
         if hasattr(generation, 'sequences') and hasattr(generation, 'audios_length'):
             audio = generation.sequences[0, :generation.audios_length[0]]
             audio_np = audio.to(torch.float32).cpu().numpy().squeeze()
@@ -224,23 +273,18 @@ def generate_base(text, description,):
                 audio_np = audio_np.flatten()
             all_audio.append(audio_np)
     
-    # Combine all audio chunks
     combined_audio = np.concatenate(all_audio)
     
-    # Convert to expected format and yield
     print(f"Sample of length: {round(combined_audio.shape[0] / sampling_rate, 2)} seconds")
     yield numpy_to_mp3(combined_audio, sampling_rate=sampling_rate)
 
-
 @spaces.GPU
-def generate_finetuned(text, description):
-    # Initialize variables
-    chunk_size = 25  # Process max 25 words or a sentence at a time
+def generate_finetuned(text, description, temperature, top_p, top_k):
+    chunk_size = 25
     
-    # Tokenize the full text and description
     inputs = description_tokenizer(description, return_tensors="pt").to(device)
 
-    sentences_text = nltk.sent_tokenize(text) # this gives us a list of sentences
+    sentences_text = nltk.sent_tokenize(text)
     curr_sentence = ""
     chunks = []
     for sentence in sentences_text:
@@ -258,22 +302,21 @@ def generate_finetuned(text, description):
     
     all_audio = []
     
-    # Process each chunk
     for chunk in chunks:
-        # Tokenize the chunk
         prompt = tokenizer(chunk, return_tensors="pt").to(device)
 
-        # Generate audio for the chunk
         generation = finetuned_model.generate(
             input_ids=inputs.input_ids,
             attention_mask=inputs.attention_mask,
             prompt_input_ids=prompt.input_ids,
             prompt_attention_mask=prompt.attention_mask,
             do_sample=True,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
             return_dict_in_generate=True
         )
 
-        # Extract audio from generation
         if hasattr(generation, 'sequences') and hasattr(generation, 'audios_length'):
             audio = generation.sequences[0, :generation.audios_length[0]]
             audio_np = audio.to(torch.float32).cpu().numpy().squeeze()
@@ -281,14 +324,11 @@ def generate_finetuned(text, description):
                 audio_np = audio_np.flatten()
             all_audio.append(audio_np)
     
-    # Combine all audio chunks
     combined_audio = np.concatenate(all_audio)
     
-    # Convert to expected format and yield
     print(f"Sample of length: {round(combined_audio.shape[0] / sampling_rate, 2)} seconds")
     yield numpy_to_mp3(combined_audio, sampling_rate=sampling_rate)
 
-
 css = """
         #share-btn-container {
             display: flex;
@@ -325,6 +365,7 @@ css = """
             display: none !important;
         }
 """
+
 with gr.Blocks(css=css) as block:
     gr.HTML(
         """
@@ -335,7 +376,7 @@ with gr.Blocks(css=css) as block:
                 "
               >
                 <h1 style="font-weight: 900; margin-bottom: 7px; line-height: normal;">
-                  Parler-TTS 🗣️
+                  Indic-Parler-TTS 🗣️
                 </h1>
               </div>
             </div>
@@ -362,11 +403,18 @@ with gr.Blocks(css=css) as block:
             with gr.Column():
                 input_text = gr.Textbox(label="Input Text", lines=2, value=finetuned_examples[0][0], elem_id="input_text")
                 description = gr.Textbox(label="Description", lines=2, value=finetuned_examples[0][1], elem_id="input_description")
+                
+                with gr.Row():
+                    temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.8, step=0.1, label="Temperature", info="Controls randomness in generation (higher = more random)")
+                    top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.1, label="Top P", info="Nucleus sampling threshold")
+                    top_k = gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Top K", info="Number of highest probability tokens to consider")
+                
                 run_button = gr.Button("Generate Audio", variant="primary")
+            
             with gr.Column():
                 audio_out = gr.Audio(label="Parler-TTS generation", format="mp3", elem_id="audio_out", autoplay=True)
 
-        inputs = [input_text, description]
+        inputs = [input_text, description, temperature, top_p, top_k]
         outputs = [audio_out]
         gr.Examples(examples=finetuned_examples, fn=generate_finetuned, inputs=inputs, outputs=outputs, cache_examples=False)
         run_button.click(fn=generate_finetuned, inputs=inputs, outputs=outputs, queue=True)
@@ -376,20 +424,26 @@ with gr.Blocks(css=css) as block:
             with gr.Column():
                 input_text = gr.Textbox(label="Input Text", lines=2, value=default_text, elem_id="input_text")
                 description = gr.Textbox(label="Description", lines=2, value="", elem_id="input_description")
+                
+                with gr.Row():
+                    temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.8, step=0.1, label="Temperature", info="Controls randomness in generation (higher = more random)")
+                    top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.1, label="Top P", info="Nucleus sampling threshold")
+                    top_k = gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Top K", info="Number of highest probability tokens to consider")
+                
                 run_button = gr.Button("Generate Audio", variant="primary")
+            
             with gr.Column():
                 audio_out = gr.Audio(label="Parler-TTS generation", format="mp3", elem_id="audio_out", autoplay=True)
 
-        inputs = [input_text, description]
+        inputs = [input_text, description, temperature, top_p, top_k]
         outputs = [audio_out]
         gr.Examples(examples=examples, fn=generate_base, inputs=inputs, outputs=outputs, cache_examples=False)
         run_button.click(fn=generate_base, inputs=inputs, outputs=outputs, queue=True)
 
-
     gr.HTML(
         """
         If you'd like to learn more about how the model was trained or explore fine-tuning it yourself, visit the <a href="https://github.com/huggingface/parler-tts">Parler-TTS</a> repository on GitHub. The Parler-TTS codebase and associated checkpoints are licensed under the <a href="https://github.com/huggingface/parler-tts/blob/main/LICENSE">Apache 2.0 license</a>.</p>
-                """
+        """
     )
 
 block.queue()