rename

transcode/silence.go

@@ -0,0 +1,112 @@
+package transcode
+
+import (
+	"bufio"
+	"fmt"
+	"os/exec"
+	"regexp"
+	"strconv"
+)
+
+// ──────────────────────────────────────────────────────────────────────────────
+// Silence detection
+// ──────────────────────────────────────────────────────────────────────────────
+
+// DetectSpeechSegments runs ffmpeg's silencedetect filter and returns the
+// non-silent (speech) segments. A "segments" WebSocket message is broadcast
+// when detection finishes.
+func DetectSpeechSegments(inputPath string, noiseDb, minDuration, padding float64, broadcast func([]byte)) ([]Segment, error) {
+	filter := fmt.Sprintf("silencedetect=noise=%.0fdB:d=%.2f", noiseDb, minDuration)
+
+	cmd := exec.Command("ffmpeg",
+		"-i", inputPath,
+		"-af", filter,
+		"-f", "null", "-",
+	)
+
+	// ffmpeg writes silencedetect output to stderr
+	stderr, err := cmd.StderrPipe()
+	if err != nil {
+		return nil, err
+	}
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("ffmpeg start: %w", err)
+	}
+
+	var (
+		silences      []silenceInterval
+		pendingStart  float64
+		totalDuration float64
+
+		durationRe = regexp.MustCompile(`Duration:\s+(\d+):(\d+):([0-9.]+)`)
+		startRe    = regexp.MustCompile(`silence_start:\s*([0-9.e+\-]+)`)
+		endRe      = regexp.MustCompile(`silence_end:\s*([0-9.e+\-]+)`)
+	)
+
+	scanner := bufio.NewScanner(stderr)
+	for scanner.Scan() {
+		line := scanner.Text()
+
+		if m := durationRe.FindStringSubmatch(line); m != nil {
+			h, _ := strconv.ParseFloat(m[1], 64)
+			min, _ := strconv.ParseFloat(m[2], 64)
+			s, _ := strconv.ParseFloat(m[3], 64)
+			totalDuration = h*3600 + min*60 + s
+		}
+		if m := startRe.FindStringSubmatch(line); m != nil {
+			pendingStart, _ = strconv.ParseFloat(m[1], 64)
+		}
+		if m := endRe.FindStringSubmatch(line); m != nil {
+			end, _ := strconv.ParseFloat(m[1], 64)
+			silences = append(silences, silenceInterval{start: pendingStart, end: end})
+		}
+	}
+
+	if err := cmd.Wait(); err != nil {
+		return nil, fmt.Errorf("ffmpeg: %w", err)
+	}
+
+	segments := removeSilence(silences, totalDuration, padding)
+
+	send(&wsMsg{
+		Type:     "segments",
+		Segments: segments,
+		Duration: totalDuration,
+	}, broadcast)
+
+	return segments, nil
+}
+
+type silenceInterval struct{ start, end float64 }
+
+// removeSilence turns silence regions into speech regions, with a small
+// padding buffer so words at the edges don't get clipped.
+func removeSilence(silences []silenceInterval, totalDuration, padding float64) []Segment {
+	if len(silences) == 0 {
+		return []Segment{{Start: 0, End: totalDuration}}
+	}
+
+	var segments []Segment
+	cursor := 0.0
+
+	for _, s := range silences {
+		segEnd := s.start + padding
+		if segEnd > totalDuration {
+			segEnd = totalDuration
+		}
+		if segEnd-cursor > 0.05 {
+			segments = append(segments, Segment{Start: cursor, End: segEnd})
+		}
+		next := s.end - padding
+		if next < segEnd {
+			next = segEnd
+		}
+		cursor = next
+	}
+
+	if cursor < totalDuration-0.01 {
+		segments = append(segments, Segment{Start: cursor, End: totalDuration})
+	}
+
+	return segments
+}