Add Kotlin and Java API for FireRedAsr AED model (#1870)

This commit is contained in:
Fangjun Kuang
2025-02-17 10:50:25 +08:00
committed by GitHub
parent 316424b382
commit d148860d2c
10 changed files with 188 additions and 1 deletions

View File

@@ -26,6 +26,7 @@ java_files += OnlineRecognizer.java
java_files += OfflineTransducerModelConfig.java
java_files += OfflineParaformerModelConfig.java
java_files += OfflineWhisperModelConfig.java
java_files += OfflineFireRedAsrModelConfig.java
java_files += OfflineMoonshineModelConfig.java
java_files += OfflineNemoEncDecCtcModelConfig.java
java_files += OfflineSenseVoiceModelConfig.java

View File

@@ -0,0 +1,42 @@
package com.k2fsa.sherpa.onnx;
public class OfflineFireRedAsrModelConfig {
private final String encoder;
private final String decoder;
private OfflineFireRedAsrModelConfig(Builder builder) {
this.encoder = builder.encoder;
this.decoder = builder.decoder;
}
public static Builder builder() {
return new Builder();
}
public String getEncoder() {
return encoder;
}
public String getDecoder() {
return decoder;
}
public static class Builder {
private String encoder = "";
private String decoder = "";
public OfflineFireRedAsrModelConfig build() {
return new OfflineFireRedAsrModelConfig(this);
}
public Builder setEncoder(String encoder) {
this.encoder = encoder;
return this;
}
public Builder setDecoder(String decoder) {
this.decoder = decoder;
return this;
}
}
}

View File

@@ -6,6 +6,7 @@ public class OfflineModelConfig {
private final OfflineTransducerModelConfig transducer;
private final OfflineParaformerModelConfig paraformer;
private final OfflineWhisperModelConfig whisper;
private final OfflineFireRedAsrModelConfig fireRedAsr;
private final OfflineMoonshineModelConfig moonshine;
private final OfflineNemoEncDecCtcModelConfig nemo;
private final OfflineSenseVoiceModelConfig senseVoice;
@@ -23,6 +24,7 @@ public class OfflineModelConfig {
this.transducer = builder.transducer;
this.paraformer = builder.paraformer;
this.whisper = builder.whisper;
this.fireRedAsr = builder.fireRedAsr;
this.moonshine = builder.moonshine;
this.nemo = builder.nemo;
this.senseVoice = builder.senseVoice;
@@ -96,6 +98,7 @@ public class OfflineModelConfig {
private OfflineParaformerModelConfig paraformer = OfflineParaformerModelConfig.builder().build();
private OfflineTransducerModelConfig transducer = OfflineTransducerModelConfig.builder().build();
private OfflineWhisperModelConfig whisper = OfflineWhisperModelConfig.builder().build();
private OfflineFireRedAsrModelConfig fireRedAsr = OfflineFireRedAsrModelConfig.builder().build();
private OfflineMoonshineModelConfig moonshine = OfflineMoonshineModelConfig.builder().build();
private OfflineNemoEncDecCtcModelConfig nemo = OfflineNemoEncDecCtcModelConfig.builder().build();
private OfflineSenseVoiceModelConfig senseVoice = OfflineSenseVoiceModelConfig.builder().build();
@@ -137,6 +140,11 @@ public class OfflineModelConfig {
return this;
}
public Builder setFireRedAsr(OfflineFireRedAsrModelConfig fireRedAsr) {
this.fireRedAsr = fireRedAsr;
return this;
}
public Builder setSenseVoice(OfflineSenseVoiceModelConfig senseVoice) {
this.senseVoice = senseVoice;
return this;

View File

@@ -174,6 +174,26 @@ static OfflineRecognizerConfig GetOfflineConfig(JNIEnv *env, jobject config) {
ans.model_config.whisper.tail_paddings =
env->GetIntField(whisper_config, fid);
// FireRedAsr
fid = env->GetFieldID(model_config_cls, "fireRedAsr",
"Lcom/k2fsa/sherpa/onnx/OfflineFireRedAsrModelConfig;");
jobject fire_red_asr_config = env->GetObjectField(model_config, fid);
jclass fire_red_asr_config_cls = env->GetObjectClass(fire_red_asr_config);
fid =
env->GetFieldID(fire_red_asr_config_cls, "encoder", "Ljava/lang/String;");
s = (jstring)env->GetObjectField(fire_red_asr_config, fid);
p = env->GetStringUTFChars(s, nullptr);
ans.model_config.fire_red_asr.encoder = p;
env->ReleaseStringUTFChars(s, p);
fid =
env->GetFieldID(fire_red_asr_config_cls, "decoder", "Ljava/lang/String;");
s = (jstring)env->GetObjectField(fire_red_asr_config, fid);
p = env->GetStringUTFChars(s, nullptr);
ans.model_config.fire_red_asr.decoder = p;
env->ReleaseStringUTFChars(s, p);
// moonshine
fid = env->GetFieldID(model_config_cls, "moonshine",
"Lcom/k2fsa/sherpa/onnx/OfflineMoonshineModelConfig;");

View File

@@ -33,6 +33,11 @@ data class OfflineWhisperModelConfig(
var tailPaddings: Int = 1000, // Padding added at the end of the samples
)
data class OfflineFireRedAsrModelConfig(
var encoder: String = "",
var decoder: String = "",
)
data class OfflineMoonshineModelConfig(
var preprocessor: String = "",
var encoder: String = "",
@@ -50,6 +55,7 @@ data class OfflineModelConfig(
var transducer: OfflineTransducerModelConfig = OfflineTransducerModelConfig(),
var paraformer: OfflineParaformerModelConfig = OfflineParaformerModelConfig(),
var whisper: OfflineWhisperModelConfig = OfflineWhisperModelConfig(),
var fireRedAsr: OfflineFireRedAsrModelConfig = OfflineFireRedAsrModelConfig(),
var moonshine: OfflineMoonshineModelConfig = OfflineMoonshineModelConfig(),
var nemo: OfflineNemoEncDecCtcModelConfig = OfflineNemoEncDecCtcModelConfig(),
var senseVoice: OfflineSenseVoiceModelConfig = OfflineSenseVoiceModelConfig(),
@@ -464,6 +470,17 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? {
modelType = "transducer",
)
}
24 -> {
val modelDir = "sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16"
return OfflineModelConfig(
fireRedAsr = OfflineFireRedAsrModelConfig(
encoder = "$modelDir/encoder.int8.onnx",
decoder = "$modelDir/decoder.int8.onnx",
),
tokens = "$modelDir/tokens.txt",
)
}
}
return null
}