app.py

import io
import json
import os
import requests
import urllib

import gradio as gr
import numpy as np
from scipy.io import wavfile


base_url = "https://api.sandbox.deepgram.com/tts"
token_str = os.environ['DG_TOKEN']
def tts_fn(text, prompt_audio, prompt_seconds, inference_steps, inference_temperature, pitch_steps):
    texts = [text]
    sr = prompt_audio[0]
    prompt_audio = np.reshape(prompt_audio[1], (1, 1, -1)).astype(np.float32, order='C') / 32768.0
    params={'synthesize': 'true', 'pitch_steps': int(pitch_steps), 'soundstorm_steps': inference_steps, 'temperature': inference_temperature, 'prompt_seconds': prompt_seconds}
    files=[('texts', ('texts', json.dumps(texts), 'application/json')), ('prompt_audio', ('prompt_audio', json.dumps(prompt_audio.tolist()), 'application/json'))]
    response = requests.post(base_url, files=files, params=params, headers={'Authorization': f'Token {token_str}'}).json()
    try:
        sample_rate = int(response['results'][0]['sample_rate'])
        audio = (np.array(response['results'][0]['audio']).transpose() / 1.414 * 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/man_3.wav', 'demo_files/woman_3.wav', 'demo_files/woman_4.wav', 'demo_files/meditation.wav']


app = gr.Blocks()

with app:
    with gr.Tab("TTS MVP"):
        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 = "Your request has been processed and the audio is ready for playback."
                textbox = gr.TextArea(label="Text", placeholder="Type a sentence here", value=cherry)
                prompt_audio = gr.Audio(label="Prompt Audio", 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")
                # 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")
                prompt_seconds = gr.Slider(minimum=1.0, maximum=10.0, value=3.0, step=1.0, label="Use first N seconds of prompt audio")
                pitch_steps = gr.Slider(minimum=-24, maximum=24, value=0, step=1, label="Pitch Steps: 12 to an octave")

            with gr.Column():
                audio_output = gr.Audio(label="Output Audio", elem_id='tts-audio')
                btn = gr.Button("Generate")
                btn.click(tts_fn, inputs=[textbox, prompt_audio, prompt_seconds, inference_steps, inference_temperature, pitch_steps], outputs=[audio_output])
app.launch()