Prompting changes

app.py

@@ -8,20 +8,25 @@ import numpy as np
 
 base_url = "https://api.sandbox.deepgram.com/nlu"
 token_str = os.environ['DG_TOKEN']
-def tts_fn(text, speed, pitch_steps, variability):
+def tts_fn(text, prompt_audio, pitch_steps, inference_steps, inference_temperature):
     texts = [text]
+    prompt_audio = np.reshape(prompt_audio[1], (1, 1, -1)).astype(np.float32, order='C') / 32768.0
     response = requests.post(
         f'{base_url}', 
-        files=[('texts', ('texts', json.dumps(texts), 'application/json'))], 
-        params={'synthesize': 'true', 'speed': speed, 'pitch_steps': int(pitch_steps), 'variability': variability},
+        files=[('texts', ('texts', json.dumps(texts), 'application/json')), ('prompt_audio', ('prompt_audio', json.dumps(prompt_audio.tolist()), 'application/json'))], 
+        params={'synthesize': 'true', 'pitch_steps': int(pitch_steps), 'soundstorm_steps': inference_steps, 'temperature': inference_temperature},
         headers={
             'Authorization': f'Token {token_str}'
         },
     ).json()
-    sample_rate = int(response['results'][0]['sample_rate'])
-    audio = np.array(response['results'][0]['audio'], dtype=np.float32)
+    try:
+        sample_rate = int(response['results'][0]['sample_rate'])
+        audio = (np.array(response['results'][0]['audio']).transpose() * 32767).astype(np.int16)
+    except Exception:
+        print(response)
     return (sample_rate, audio)
 
+demo_files = ['demo_files/man.wav', 'demo_files/woman.wav', 'demo_files/man_2.wav', 'demo_files/woman_2.wav', 'demo_files/meditation.wav']
 
 app = gr.Blocks()
 
@@ -30,14 +35,18 @@ with app:
         with gr.Row():
             with gr.Column():
                 pangram = "The beige hue on the waters of the loch impressed all, including the French queen, before she heard that symphony again, just as young Arthur wanted."
-                cherry = "Good evening. I'm a text to speech bot. Please allow me to assist you."
+                cherry = "Your request has been processed and the audio is ready for playback."
                 textbox = gr.TextArea(label="Text", placeholder="Type a sentence here", value=cherry)
-                speed = gr.Slider(minimum=0.0, maximum=2.0, value=1.1, step=0.1, label="Speed")
+                prompt_audio = gr.Audio(label="Prompt Audio (first 3 seconds of selection)", source='upload')
+                examples = gr.Examples(label='Sample Speakers', examples=demo_files, inputs=prompt_audio)
+                # speed = gr.Slider(minimum=0.0, maximum=2.0, value=1.1, step=0.1, label="Speed")
                 pitch_steps = gr.Slider(minimum=-24, maximum=24, value=0, step=1, label="Pitch Steps: 12 to an octave")
-                variability = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, label="Variability")
+                # variability = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, label="Variability")
+                inference_steps = gr.Slider(minimum=1, maximum=32, value=1, step=1, label="Inference Steps: quality vs latency tradeoff. Results are sometimes unstable for values >1.")
+                inference_temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.9, step=0.05, label="Temperature: fidelity vs variability tradeoff")
 
             with gr.Column():
                 audio_output = gr.Audio(label="Output Audio", elem_id='tts-audio')
                 btn = gr.Button("Generate")
-                btn.click(tts_fn, inputs=[textbox, speed, pitch_steps, variability], outputs=[audio_output])
-app.launch()
+                btn.click(tts_fn, inputs=[textbox, prompt_audio, pitch_steps, inference_steps, inference_temperature], outputs=[audio_output])
+app.launch(share=True)