AI Music and Audio Generation System Development

Development of AI-based music and audio generation systems

AI music generation automates the creation of background music, jingles, and sound effects for content, games, and advertising. It replaces stock music licensing and studio recording for those with basic requirements.

Platform Comparison

AudioCraft / MusicGen — self-hosted

from audiocraft.models import MusicGen
from audiocraft.data.audio import audio_write
import torch

class MusicGenerator:
    def __init__(self, model_size: str = "medium"):
        # Размеры: small (300M), medium (1.5B), large (3.3B), melody
        self.model = MusicGen.get_pretrained(f"facebook/musicgen-{model_size}")
        self.model.set_generation_params(
            duration=30,        # секунды (max 30 для standard, до 120 через chunking)
            temperature=1.0,    # 0.5–1.5
            top_k=250,
            top_p=0.0,
            cfg_coef=3.0        # adherence to prompt
        )

    def generate(
        self,
        description: str,
        duration: int = 30,
        temperature: float = 1.0
    ) -> bytes:
        self.model.set_generation_params(duration=duration, temperature=temperature)

        wav = self.model.generate(
            descriptions=[description],
            progress=True
        )

        import io
        import torchaudio
        buf = io.BytesIO()
        torchaudio.save(buf, wav[0].cpu(), sample_rate=32000, format="mp3")
        return buf.getvalue()

    def generate_with_melody(
        self,
        description: str,
        melody_audio: bytes,
        duration: int = 30
    ) -> bytes:
        """Генерируем музыку по мотивам референсной мелодии"""
        import io
        import torchaudio
        melody_wav, sr = torchaudio.load(io.BytesIO(melody_audio))

        model = MusicGen.get_pretrained("facebook/musicgen-melody")
        model.set_generation_params(duration=duration)

        wav = model.generate_with_chroma(
            descriptions=[description],
            melody_wavs=melody_wav.unsqueeze(0),
            melody_sample_rate=sr,
            progress=True
        )

        buf = io.BytesIO()
        torchaudio.save(buf, wav[0].cpu(), sample_rate=32000, format="mp3")
        return buf.getvalue()

Sound Effect Generation (AudioGen)

from audiocraft.models import AudioGen

sfx_model = AudioGen.get_pretrained("facebook/audiogen-medium")
sfx_model.set_generation_params(duration=5)

def generate_sound_effect(description: str, duration: float = 3.0) -> bytes:
    sfx_model.set_generation_params(duration=duration)
    wav = sfx_model.generate(descriptions=[description])

    import io, torchaudio
    buf = io.BytesIO()
    torchaudio.save(buf, wav[0].cpu(), sample_rate=16000, format="wav")
    return buf.getvalue()

# Примеры: "forest ambience with birds", "robot beeping", "door creaking"

Contextual Applications

FastAPI service

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()
music_gen = MusicGenerator("medium")

class MusicRequest(BaseModel):
    description: str
    duration: int = 30
    temperature: float = 1.0

@app.post("/generate/music")
async def generate_music(req: MusicRequest):
    audio = music_gen.generate(req.description, req.duration, req.temperature)
    return Response(content=audio, media_type="audio/mpeg")

Delivery time: Self-hosted MusicGen API – 1–2 days. Platform with multiple models, queue, and CDN storage – 2–3 weeks.