media.mjs

import { BUFFER, convert, FILE, getTempPath } from './storage.mjs';
import { hash } from './encryption.mjs';
import { need } from './utilitas.mjs';

const _NEED = [
    'fluent-ffmpeg',
    '@ffmpeg-installer/ffmpeg',
    '@ffprobe-installer/ffprobe',
];

const writeStringToBuffer = (buffer, offset, string) => {
    for (let i = 0; i < string.length; i++) {
        buffer.writeUInt8(string.charCodeAt(i), offset + i);
    }
};

const createWavHeader = (
    dataSize, sampleRate = 24000, // 16kHz sample rate
    numChannels = 1, bitsPerSample = 16
) => {
    const header = Buffer.alloc(44);
    // 'RIFF' chunk descriptor
    writeStringToBuffer(header, 0, 'RIFF');
    header.writeUInt32LE(36 + dataSize, 4); // File size - 8
    writeStringToBuffer(header, 8, 'WAVE');
    // 'fmt ' sub-chunk
    writeStringToBuffer(header, 12, 'fmt ');
    header.writeUInt32LE(16, 16); // Sub-chunk size (16 for PCM)
    header.writeUInt16LE(1, 20);  // Audio format (1 for PCM)
    header.writeUInt16LE(numChannels, 22); // Number of channels
    header.writeUInt32LE(sampleRate, 24);  // Sample rate
    header.writeUInt32LE(sampleRate * numChannels * bitsPerSample / 8, 28); // Byte rate
    header.writeUInt16LE(numChannels * bitsPerSample / 8, 32); // Block align
    header.writeUInt16LE(bitsPerSample, 34); // Bits per sample
    // 'data' sub-chunk
    writeStringToBuffer(header, 36, 'data');
    header.writeUInt32LE(dataSize, 40); // Data size
    return header;
};

// https://codex.so/ffmpeg-node-js
const getFfmpeg = async (options) => {
    const ffmpeg = await need('fluent-ffmpeg');
    ffmpeg.setFfmpegPath((await need('@ffmpeg-installer/ffmpeg')).path);
    ffmpeg.setFfprobePath((await need('@ffprobe-installer/ffprobe')).path);
    return new ffmpeg(options);
};

const convertAudio = async (input, options) => {
    assert(input, 'Invalid audio input.', 400);
    const { content, cleanup } = await convert(input, {
        ...options || {}, expected: FILE, withCleanupFunc: true,
    });
    const conv = await getFfmpeg(content);
    const output = options?.output || getTempPath({
        sub: options?.suffix
            ? `${hash(JSON.stringify({ content, options }))}.${options.suffix}`
            : null,
    });
    options?.frequency && conv.audioFrequency(options.frequency);
    options?.channels && conv.audioChannels(options.channels);
    options?.codec && conv.audioCodec(options?.codec);
    await new Promise((resolve, reject) => {
        conv.on('error', reject); // (err, stdout, stderr)
        conv.on('end', resolve);
        conv.save(output);
    });
    options?.input === FILE || await cleanup();
    return await convert(output, {
        expected: BUFFER, cleanup: true, ...options || {}, input: FILE,
    });
};

// Suitable for Whisper AI models: https://github.com/ggerganov/whisper.cpp
// > ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
const convertAudioTo16kNanoPcmWave = (input, options) => convertAudio(input, {
    channels: 1, codec: 'pcm_s16le', frequency: 16000, suffix: 'wav',
    ...options || {},
});

// Suitable for Telegram Voice: https://core.telegram.org/bots/api#sendvoice
// https://stackoverflow.com/questions/44615991/how-convert-ogg-file-to-telegram-voice-format
// ffmpeg -i input.mp3 -vn -acodec libopus -b:a 16k audio.ogg
const convertAudioTo16kNanoOpusOgg = (input, options) => convertAudio(input, {
    channels: 1, codec: 'libopus', frequency: 48000, suffix: 'ogg',
    ...options || {},
});

export {
    _NEED,
    convertAudioTo16kNanoOpusOgg,
    convertAudioTo16kNanoPcmWave,
    createWavHeader,
    getFfmpeg,
};