go-whisper-api/transcode/aac_decode.go

package transcode

import (
    "fmt"
    "io"
    "math"
    "os"
    "path/filepath"
    "strings"

    "github.com/olivier-w/climp-aac-decoder/aacfile"
)

func decodeAACPath(path, ext string) ([]float64, int, int, error) {
    f, err := os.Open(path)
    if err != nil {
        return nil, 0, 0, err
    }
    defer f.Close()
    st, err := f.Stat()
    if err != nil {
        return nil, 0, 0, err
    }
    // aacfile picks the parser from the *name* extension, not file content (.mp4 → .m4a).
    name := aacOpenName(path, ext)
    size := st.Size()
    r, err := aacfile.Open(f, size, name)
    if err != nil && isMP4SampleDeltaError(err) {
        return decodeMP4AACRelaxed(f, size)
    }
    if err != nil {
        return nil, 0, 0, err
    }
    defer r.Close()
    sr := r.SampleRate()
    ch := r.ChannelCount()
    pcm, err := io.ReadAll(r)
    if err != nil {
        return nil, 0, 0, fmt.Errorf("read aac pcm: %w", err)
    }
    samples := pcm16LEToFloat(pcm, ch)
    if ch > 1 {
        samples = interleavedToMono(samples, ch)
        ch = 1
    }
    return samples, sr, ch, nil
}

// aacContainerExt maps file extensions to a container name understood by aacfile.
func aacContainerExt(ext string) string {
    switch strings.ToLower(ext) {
    case ".mp4", ".m4v", ".mov", ".3gp", ".3g2":
        return ".m4a"
    case ".aac", ".m4a", ".m4b":
        return ext
    default:
        return ".m4a"
    }
}

func aacOpenName(path, ext string) string {
    containerExt := aacContainerExt(ext)
    if containerExt == "" {
        containerExt = ".m4a"
    }
    base := filepath.Base(path)
    if e := strings.ToLower(filepath.Ext(base)); e == containerExt {
        return base
    }
    stem := strings.TrimSuffix(base, filepath.Ext(base))
    if stem == "" || stem == base {
        stem = "audio"
    }
    return stem + containerExt
}

func pcm16LEToFloat(pcm []byte, channels int) []float64 {
    if channels <= 0 {
        channels = 1
    }
    frameBytes := 2 * channels
    nFrames := len(pcm) / frameBytes
    out := make([]float64, nFrames*channels)
    for i := 0; i < nFrames*channels; i++ {
        off := i * 2
        if off+1 >= len(pcm) {
            break
        }
        v := int16(pcm[off]) | int16(pcm[off+1])<<8
        out[i] = float64(v) / 32768.0
    }
    return out
}

func interleavedToMono(samples []float64, channels int) []float64 {
    if channels <= 1 {
        return samples
    }
    nFrames := len(samples) / channels
    out := make([]float64, nFrames)
    for i := 0; i < nFrames; i++ {
        var sum float64
        for c := 0; c < channels; c++ {
            sum += samples[i*channels+c]
        }
        out[i] = sum / float64(channels)
    }
    return out
}

func resampleLinear(samples []float64, fromRate, toRate int) []float64 {
    if fromRate <= 0 || toRate <= 0 || fromRate == toRate || len(samples) == 0 {
        return samples
    }
    ratio := float64(fromRate) / float64(toRate)
    outLen := int(math.Round(float64(len(samples)) / ratio))
    if outLen < 1 {
        outLen = 1
    }
    out := make([]float64, outLen)
    for i := 0; i < outLen; i++ {
        src := float64(i) * ratio
        j := int(src)
        if j >= len(samples)-1 {
            out[i] = samples[len(samples)-1]
            continue
        }
        frac := src - float64(j)
        out[i] = samples[j]*(1-frac) + samples[j+1]*frac
    }
    return out
}