Add Dart API for keyword spotter (#1162)

.github/scripts/test-dart.sh

@@ -4,6 +4,10 @@ set -ex
 
 cd dart-api-examples
 
+pushd keyword-spotter
+./run-zh.sh
+popd
+
 pushd non-streaming-asr
 
 echo '----------SenseVoice----------'

.github/workflows/test-dart.yaml

@@ -108,6 +108,7 @@ jobs:
           cp scripts/dart/non-streaming-asr-pubspec.yaml dart-api-examples/non-streaming-asr/pubspec.yaml
           cp scripts/dart/streaming-asr-pubspec.yaml dart-api-examples/streaming-asr/pubspec.yaml
           cp scripts/dart/tts-pubspec.yaml dart-api-examples/tts/pubspec.yaml
+          cp scripts/dart/kws-pubspec.yaml dart-api-examples/keyword-spotter/pubspec.yaml
           cp scripts/dart/sherpa-onnx-pubspec.yaml flutter/sherpa_onnx/pubspec.yaml
 
           .github/scripts/test-dart.sh

CHANGELOG.md

@@ -1,6 +1,7 @@
 ## 1.10.17
 
 * Support SenseVoice CTC models.
+* Add Dart API for keyword spotter.
 
 ## 1.10.16
 

dart-api-examples/.gitignore

@@ -1 +1,28 @@
 !run*.sh
+# See https://www.dartlang.org/guides/libraries/private-files
+
+# Files and directories created by pub
+.dart_tool/
+.packages
+build/
+# If you're building an application, you may want to check-in your pubspec.lock
+pubspec.lock
+
+# Directory created by dartdoc
+# If you don't generate documentation locally you can remove this line.
+doc/api/
+
+# dotenv environment variables file
+.env*
+
+# Avoid committing generated Javascript files:
+*.dart.js
+*.info.json      # Produced by the --dump-info flag.
+*.js             # When generated by dart2js. Don't specify *.js if your
+                 # project includes source files written in JavaScript.
+*.js_
+*.js.deps
+*.js.map
+
+.flutter-plugins
+.flutter-plugins-dependencies

dart-api-examples/keyword-spotter/.gitignore

@@ -0,0 +1,3 @@
+# https://dart.dev/guides/libraries/private-files
+# Created by `dart pub`
+.dart_tool/

dart-api-examples/keyword-spotter/CHANGELOG.md

@@ -0,0 +1,3 @@
+## 1.0.0
+
+- Initial version.

dart-api-examples/keyword-spotter/README.md

@@ -0,0 +1,4 @@
+# Introduction
+
+This directory contains keyword spotting examples using
+Dart API from [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx)

dart-api-examples/keyword-spotter/analysis_options.yaml

@@ -0,0 +1,30 @@
+# This file configures the static analysis results for your project (errors,
+# warnings, and lints).
+#
+# This enables the 'recommended' set of lints from `package:lints`.
+# This set helps identify many issues that may lead to problems when running
+# or consuming Dart code, and enforces writing Dart using a single, idiomatic
+# style and format.
+#
+# If you want a smaller set of lints you can change this to specify
+# 'package:lints/core.yaml'. These are just the most critical lints
+# (the recommended set includes the core lints).
+# The core lints are also what is used by pub.dev for scoring packages.
+
+include: package:lints/recommended.yaml
+
+# Uncomment the following section to specify additional rules.
+
+# linter:
+#   rules:
+#     - camel_case_types
+
+# analyzer:
+#   exclude:
+#     - path/to/excluded/files/**
+
+# For more information about the core and recommended set of lints, see
+# https://dart.dev/go/core-lints
+
+# For additional information about configuring this file, see
+# https://dart.dev/guides/language/analysis-options

dart-api-examples/keyword-spotter/bin/init.dart

@@ -0,0 +1 @@
+../../vad/bin/init.dart

dart-api-examples/keyword-spotter/bin/zipformer-transducer.dart

@@ -0,0 +1,98 @@
+// Copyright (c)  2024  Xiaomi Corporation
+import 'dart:io';
+import 'dart:typed_data';
+
+import 'package:args/args.dart';
+import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
+
+import './init.dart';
+
+void main(List<String> arguments) async {
+  await initSherpaOnnx();
+
+  final parser = ArgParser()
+    ..addOption('encoder', help: 'Path to the encoder model')
+    ..addOption('decoder', help: 'Path to decoder model')
+    ..addOption('joiner', help: 'Path to joiner model')
+    ..addOption('tokens', help: 'Path to tokens.txt')
+    ..addOption('keywords-file', help: 'Path to keywords.txt')
+    ..addOption('input-wav', help: 'Path to input.wav to transcribe');
+
+  final res = parser.parse(arguments);
+  if (res['encoder'] == null ||
+      res['decoder'] == null ||
+      res['joiner'] == null ||
+      res['tokens'] == null ||
+      res['keywords-file'] == null ||
+      res['input-wav'] == null) {
+    print(parser.usage);
+    exit(1);
+  }
+
+  final encoder = res['encoder'] as String;
+  final decoder = res['decoder'] as String;
+  final joiner = res['joiner'] as String;
+  final tokens = res['tokens'] as String;
+  final keywordsFile = res['keywords-file'] as String;
+  final inputWav = res['input-wav'] as String;
+
+  final transducer = sherpa_onnx.OnlineTransducerModelConfig(
+    encoder: encoder,
+    decoder: decoder,
+    joiner: joiner,
+  );
+
+  final modelConfig = sherpa_onnx.OnlineModelConfig(
+    transducer: transducer,
+    tokens: tokens,
+    debug: true,
+    numThreads: 1,
+  );
+  final config = sherpa_onnx.KeywordSpotterConfig(
+    model: modelConfig,
+    keywordsFile: keywordsFile,
+  );
+  final spotter = sherpa_onnx.KeywordSpotter(config);
+
+  final waveData = sherpa_onnx.readWave(inputWav);
+  var stream = spotter.createStream();
+
+  // simulate streaming. You can choose an arbitrary chunk size.
+  // chunkSize of a single sample is also ok, i.e, chunkSize = 1
+  final chunkSize = 1600; // 0.1 second for 16kHz
+  final numChunks = waveData.samples.length ~/ chunkSize;
+
+  for (int i = 0; i != numChunks; ++i) {
+    int start = i * chunkSize;
+    stream.acceptWaveform(
+      samples:
+          Float32List.sublistView(waveData.samples, start, start + chunkSize),
+      sampleRate: waveData.sampleRate,
+    );
+    while (spotter.isReady(stream)) {
+      spotter.decode(stream);
+      final result = spotter.getResult(stream);
+      if (result.keyword != '') {
+        print('Detected: ${result.keyword}');
+      }
+    }
+  }
+
+  // 0.5 seconds, assume sampleRate is 16kHz
+  final tailPaddings = Float32List(8000);
+  stream.acceptWaveform(
+    samples: tailPaddings,
+    sampleRate: waveData.sampleRate,
+  );
+
+  while (spotter.isReady(stream)) {
+    spotter.decode(stream);
+    final result = spotter.getResult(stream);
+    if (result.keyword != '') {
+      print('Detected: ${result.keyword}');
+    }
+  }
+
+  stream.free();
+  spotter.free();
+}

dart-api-examples/keyword-spotter/pubspec.yaml

@@ -0,0 +1,19 @@
+name: keyword_spotter
+
+description: >
+  This example demonstrates how to use the Dart API for keyword spotting
+
+version: 1.0.0
+
+environment:
+  sdk: ^3.4.0
+
+dependencies:
+  sherpa_onnx: ^1.10.17
+  # sherpa_onnx:
+  #   path: ../../flutter/sherpa_onnx
+  path: ^1.9.0
+  args: ^2.5.0
+
+dev_dependencies:
+  lints: ^3.0.0

dart-api-examples/keyword-spotter/run-zh.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+
+set -ex
+
+dart pub get
+
+if [ ! -f ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/tokens.txt ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/kws-models/sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01.tar.bz2
+  tar xvf sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01.tar.bz2
+  rm sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01.tar.bz2
+fi
+
+dart run \
+  ./bin/zipformer-transducer.dart \
+  --encoder ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/encoder-epoch-12-avg-2-chunk-16-left-64.onnx \
+  --decoder ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/decoder-epoch-12-avg-2-chunk-16-left-64.onnx \
+  --joiner ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/joiner-epoch-12-avg-2-chunk-16-left-64.onnx \
+  --tokens ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/tokens.txt \
+  --keywords-file ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/test_wavs/test_keywords.txt \
+  --input-wav ./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/test_wavs/3.wav
+

flutter/sherpa_onnx/lib/sherpa_onnx.dart

@@ -3,6 +3,7 @@ import 'dart:io';
 import 'dart:ffi';
 
 export 'src/feature_config.dart';
+export 'src/keyword_spotter.dart';
 export 'src/offline_recognizer.dart';
 export 'src/offline_stream.dart';
 export 'src/online_recognizer.dart';