handler.py

from typing import Dict, List, Any
from scipy.io import wavfile
from transformers import AutoProcessor, MusicgenForConditionalGeneration
import torch
import io
import base64
import wave
import array
import math

def generate_sine_wave(freq, duration, sample_rate, amplitude):
    n_samples = int(sample_rate * duration)
    samples = []
    
    for x in range(n_samples):
        value = amplitude * math.sin(2 * math.pi * freq * x / sample_rate)
        samples.append(int(value))  # rounding to the nearest integer
    
    return array.array("h", samples)  # array of short integers


def sine_to_base64():
    frequency = 440.0  # Frequency in Hz
    duration = 1.0     # seconds
    volume = 0.5       # 0.0 to 1.0
    sample_rate = 44100
    amplitude = int(volume * 32767)  # 16-bit audio
    
    sine_wave = generate_sine_wave(frequency, duration, sample_rate, amplitude)
    
    wav_buffer = io.BytesIO()
    with wave.open(wav_buffer, "w") as wav_file:
        n_channels = 1
        sampwidth = 2
        n_frames = len(sine_wave)
        comptype = "NONE"
        compname = "not compressed"
        wav_file.setparams((n_channels, sampwidth, int(sample_rate), n_frames, comptype, compname))
        wav_file.writeframes(sine_wave.tobytes())
    
    base64_string = base64.b64encode(wav_buffer.getvalue()).decode('utf-8')
    return base64_string


def create_params(params, fr):
    # default
    out = { "do_sample": True,
          "guidance_scale": 3, 
          "max_new_tokens": 256
          }

    has_tokens = False

    if params is None:
       return out

    if 'duration' in params:
        out['max_new_tokens'] =  params['duration'] * fr
        has_tokens = True

    for k, p in params.items():
        if k in out: 
          if has_tokens and k == 'max_new_tokens':
            continue

          out[k] = p

    return out


class EndpointHandler:
    def __init__(self, path="pbotsaris/musicgen-small"):
        # load model and processor
        self.processor = AutoProcessor.from_pretrained(path)
        self.model = MusicgenForConditionalGeneration.from_pretrained(path, torch_dtype=torch.float16)
        self.model.to('cuda')

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, str]]:
        """
        Args:
            data (:dict:):
                The payload with the text prompt and generation parameters.

       Returns: wav file in bytes
        """

        # inputs = data.pop("inputs", data)
        # params = data.pop("parameters", None)

        # inputs = self.processor(
        #     text=[inputs],
        #     padding=True,
        #     return_tensors="pt"
        # ).to('cuda')

        # params = create_params(params, self.model.config.audio_encoder.frame_rate)

        # with torch.cuda.amp.autocast():
        #     outputs = self.model.generate(**inputs, **params)

        # pred = outputs[0, 0].cpu().numpy()
        # sr = 32000

        # try:
        #      sr = self.model.config.audio_encoder.sampling_rate

        # except:
        #      sr = 32000

        # wav_buffer = io.BytesIO()
        # wavfile.write(wav_buffer, rate=sr, data=pred)
        # wav_data = wav_buffer.getvalue()

        # base64_encoded_wav = base64.b64encode(wav_data).decode('utf-8')

        base64_encoded_wav = sine_to_base64()
        return  [{"audio": base64_encoded_wav}]


if __name__ == "__main__":
    handler = EndpointHandler()