@jjhbw/silero-vad 1.0.2 → 1.0.3

package/README.md

@@ -1,4 +1,4 @@
-# Silero VAD Node Fork
+# Silero VAD Node
 
 Minimal Node.js wrapper around the Silero VAD ONNX model, with a small CLI and parity tests against the Python implementation. The Node implementation runs VAD and silence stripping directly from ffmpeg streams to keep memory usage low on long files.
 
@@ -27,9 +27,7 @@ const {
     const inputs = ["input.wav", "other.mp3"];
     for (const inputPath of inputs) {
       vad.resetStates(); // per file/stream
-      const ts = await getSpeechTimestamps(inputPath, vad, {
-        returnSeconds: true,
-      });
+      const ts = await getSpeechTimestamps(inputPath, vad);
       // Each entry includes both seconds (start/end) and samples (startSample/endSample).
       console.log(inputPath, ts);
       // Example return value:
@@ -73,7 +71,6 @@ Options:
 - `--speech-pad-ms <ms>`: padding added to each speech segment in ms (default `30`).
 - `--time-resolution <n>`: decimal places for seconds output (default `3`).
 - `--neg-threshold <float>`: override the negative threshold (default `max(threshold - 0.15, 0.01)`).
-- `--seconds`: output timestamps in seconds (default on).
 - `--cps <float>`: enable the timeline visualization and set chars per second (default `4`).
 - `--strip-silence`: write a new WAV file with silences removed.
 - `--output-dir <path>`: output directory for strip-silence files (default: input dir).

package/cli.js

@@ -3,6 +3,7 @@
 const fs = require('fs');
 const fsp = fs.promises;
 const path = require('path');
+const { spawn } = require('child_process');
 const {
   loadSileroVad,
   getSpeechTimestamps,
@@ -33,15 +34,13 @@ async function main() {
         // reuse session, reset stream state per file
         vad.resetStates();
         const t0 = performance.now();
-        const { timestamps, totalSamples } = await getSpeechTimestamps(audioPath, vad, {
+        const timestamps = await getSpeechTimestamps(audioPath, vad, {
           threshold: args.threshold,
           minSpeechDurationMs: args.minSpeechDurationMs,
           minSilenceDurationMs: args.minSilenceDurationMs,
           speechPadMs: args.speechPadMs,
-          returnSeconds: args.seconds,
           timeResolution: args.timeResolution,
           negThreshold: args.negThreshold,
-          returnMetadata: true,
         });
         const t1 = performance.now();
         results.push({ file: audioPath, timestamps });
@@ -49,10 +48,10 @@ async function main() {
         const mem = process.memoryUsage();
         const speechSeconds = getSpeechDurationSeconds(
           timestamps,
-          args.seconds,
+          true,
           effectiveSampleRate,
         );
-        const durationSeconds = totalSamples / effectiveSampleRate;
+        const durationSeconds = await getAudioDurationSeconds(audioPath);
         const silenceSeconds = Math.max(0, durationSeconds - speechSeconds);
         const totalForPct = durationSeconds > 0 ? durationSeconds : 1;
         const speechPct = (speechSeconds / totalForPct) * 100;
@@ -99,9 +98,9 @@ async function main() {
         }
 
         if (args.stripSilence) {
-          const segmentsSeconds = timestamps.map(({ start, end, startSeconds, endSeconds }) => ({
-            start: args.seconds ? start : startSeconds,
-            end: args.seconds ? end : endSeconds,
+          const segmentsSeconds = timestamps.map(({ start, end }) => ({
+            start,
+            end,
           }));
           if (!segmentsSeconds.length) {
             console.info(`strip_silence=skipped (no speech detected)`);
@@ -164,7 +163,6 @@ function parseArgs(argv) {
     speechPadMs: 30,
     timeResolution: 3,
     negThreshold: null,
-    seconds: true,
     charsPerSecond: 4,
     showTimeline: false,
     stripSilence: false,
@@ -220,8 +218,6 @@ function parseArgs(argv) {
         out.negThreshold = value;
       }
       i += 1;
-    } else if (arg === '--seconds') {
-      out.seconds = true;
     } else if (arg === '--cps') {
       const value = parseFloat(argv[i + 1]);
       out.showTimeline = true;
@@ -254,7 +250,6 @@ Options:
   --speech-pad-ms <ms>   Padding added to speech segments in ms (default: 30)
   --time-resolution <n>  Decimal places for seconds output (default: 3)
   --neg-threshold <f>    Negative threshold override (default: max(threshold - 0.15, 0.01))
-  --seconds              Output timestamps in seconds (default: on)
   --cps <float>          Enable timeline visualization; chars per second (default: 4)
   --strip-silence         Write a new file with all silences removed
   --output-dir <path>     Output directory for strip-silence files (default: input dir)
@@ -319,6 +314,44 @@ function getStripOutputPath(inputPath, outputDir) {
   return path.join(dir, `${base}_speech.wav`);
 }
 
+async function getAudioDurationSeconds(inputPath) {
+  const args = [
+    '-v',
+    'error',
+    // Use packet timestamps to handle containers where format duration is unreliable.
+    '-select_streams',
+    'a:0',
+    '-show_entries',
+    'packet=pts_time',
+    '-of',
+    'csv=p=0',
+    inputPath,
+  ];
+  return new Promise((resolve, reject) => {
+    const ffprobe = spawn('ffprobe', args, { stdio: ['ignore', 'pipe', 'inherit'] });
+    let output = '';
+    ffprobe.stdout.on('data', (chunk) => {
+      output += chunk.toString();
+    });
+    ffprobe.on('error', reject);
+    ffprobe.on('close', (code) => {
+      if (code !== 0) {
+        reject(new Error(`ffprobe exited with code ${code}`));
+        return;
+      }
+      const lines = output.trim().split('\n');
+      for (let i = lines.length - 1; i >= 0; i -= 1) {
+        const value = parseFloat(lines[i]);
+        if (Number.isFinite(value)) {
+          resolve(value);
+          return;
+        }
+      }
+      reject(new Error('Unable to read audio duration from ffprobe output'));
+    });
+  });
+}
+
 async function ensureUniquePath(outputPath) {
   try {
     await fsp.access(outputPath);

package/lib.js

@@ -1,5 +1,3 @@
-const fs = require('fs');
-const fsp = fs.promises;
 const path = require('path');
 const { spawn } = require('child_process');
 const ort = require('onnxruntime-node');
@@ -120,7 +118,6 @@ async function loadSileroVad(model = 'default', opts = {}) {
  *   Example: a 50 ms pause will not split a segment at 100 ms.
  * @param {number} [options.speechPadMs=30] Pad each segment on both sides, clamped
  *   to neighbors. Example: [1.000, 2.000] -> ~[0.970, 2.030].
- * @param {boolean} [options.returnSeconds=false]
  * @param {number} [options.timeResolution=3] Decimal places for seconds output.
  *   Example: timeResolution=1 turns 1.23456 into 1.2.
  * @param {number} [options.negThreshold=threshold-0.15] End speech when prob dips
@@ -128,7 +125,8 @@ async function loadSileroVad(model = 'default', opts = {}) {
  *   negThreshold=0.35 keeps speech open during brief 0.4 dips.
  *   Default clamps to >= 0.01 to avoid an always-on end condition.
  * @param {number} [options.sampleRate]
- * @param {boolean} [options.returnMetadata=false]
+ * @returns {Promise<Array<{start: number, end: number, startSample: number, endSample: number}>>}
+ *   start/end are seconds; startSample/endSample are sample indices.
  */
 async function getSpeechTimestamps(
   inputPath,
@@ -138,11 +136,9 @@ async function getSpeechTimestamps(
     minSpeechDurationMs = 250,
     minSilenceDurationMs = 100,
     speechPadMs = 30,
-    returnSeconds = false,
     timeResolution = 3,
     negThreshold,
     sampleRate,
-    returnMetadata = false,
   } = {},
 ) {
   if (!vad) {
@@ -351,23 +347,12 @@ async function getSpeechTimestamps(
   }
 
   const convertSeconds = (samples) => +(samples / sr).toFixed(timeResolution);
-  const result = returnSeconds
-    ? speeches.map(({ start, end }) => ({
-      start: convertSeconds(start),
-      end: convertSeconds(end),
-      startSample: start,
-      endSample: end,
-    }))
-    : speeches.map(({ start, end }) => ({
-      start,
-      end,
-      startSeconds: convertSeconds(start),
-      endSeconds: convertSeconds(end),
-    }));
-
-  if (returnMetadata) {
-    return { timestamps: result, totalSamples };
-  }
+  const result = speeches.map(({ start, end }) => ({
+    start: convertSeconds(start),
+    end: convertSeconds(end),
+    startSample: start,
+    endSample: end,
+  }));
 
   return result;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jjhbw/silero-vad",
-  "version": "1.0.2",
+  "version": "1.0.3",
   "description": "Node.js bindings for Silero VAD",
   "main": "index.js",
   "exports": {
@@ -17,7 +17,7 @@
   "license": "MIT",
   "repository": {
     "type": "git",
-    "url": "https://github.com/jjhbw/silero-vad.git"
+    "url": "git+https://github.com/jjhbw/silero-vad.git"
   },
   "bugs": {
     "url": "https://github.com/jjhbw/silero-vad/issues"