@pr0gramm/fluester/dist/ffmpeg.js

Node.js bindings for OpenAI's Whisper. Optimized for CPU.

import { path as ffmpegPath } from "@ffmpeg-installer/ffmpeg";
import { path as ffprobePath } from "@ffprobe-installer/ffprobe";
import ffmpeg from "fluent-ffmpeg";
ffmpeg.setFfmpegPath(ffmpegPath);
ffmpeg.setFfprobePath(ffprobePath);
/**
 * Converts a file to a file that whisper.cpp can work with.
 *
 * @param inputFile Any input file that ffmpeg supports. That means almost any audio or video file.
 * @param outputFile
 * @example
 * ```
 * const inputFile = "...";
 * const outputFile = "output.wav";
 * await convertFileToProcessableFile(inputFile, outputFile);
 * // ...
 *  await client.detectLanguage(outputFile);
 ```
 */
export async function convertFileToProcessableFile(inputFile, outputFile, _options) {
    // const signal = options?.signal;
    // Taken from the whisper.cpp docs:
    // ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
    const command = ffmpeg(inputFile)
        .audioFrequency(16000)
        .audioChannels(1)
        .audioCodec("pcm_s16le")
        .output(outputFile);
    return new Promise((resolve, reject) => {
        command.once("end", resolve);
        command.once("error", reject);
        command.run();
    });
}
// TODO: Not yet supported:
/*
async function pipeStreamToProcessableStream(
    input: Readable,
    output: Writable,
    _options?: ConvertOptions,
): Promise<void> {
    // const signal = options?.signal;

    // Taken from the whisper.cpp docs:
    // ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
    const command = ffmpeg(input)
        .audioFrequency(16000)
        .audioChannels(1)
        .audioCodec("pcm_s16le");

    return new Promise((resolve, reject) => {
        command.once("end", resolve);
        command.once("error", reject);
        command.pipe(output, { end: true });
    });
}
*/