Add Pascal API for FireRedAsr AED Model (#1877) (#1880)

2025-02-17 16:06:18 +08:00
parent 87a968b55d
commit 614c51068b
4 changed files with 198 additions and 48 deletions
--- a/.github/workflows/pascal.yaml
+++ b/.github/workflows/pascal.yaml
@@ -125,6 +125,56 @@ jobs:
            cp -v ../sherpa-onnx/pascal-api/*.pas ../pascal-api-examples/tts
          fi

+      - name:  Run Pascal test (Non Streaming ASR)
+        shell: bash
+        run: |
+          export PATH=/c/lazarus/fpc/3.2.2/bin/x86_64-win64:$PATH
+
+          cd ./pascal-api-examples
+
+          pushd non-streaming-asr
+          ./run-zipformer-transducer.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-moonshine.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-fire-red-asr.sh
+          rm -rf sherpa-onnx-fire-red-asr*
+          echo "---"
+
+          ./run-whisper.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-nemo-transducer.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-nemo-ctc.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-sense-voice.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-telespeech-ctc.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ./run-paraformer.sh
+
+          ./run-paraformer-itn.sh
+
+          rm -rf sherpa-onnx-*
+          echo "---"
+
+          ls -lh
+          popd
+
      - name:  Run Pascal test (Speaker diarization)
        shell: bash
        run: |
@@ -235,52 +285,6 @@ jobs:
          ls -lh
          popd

-      - name:  Run Pascal test (Non Streaming ASR)
-        shell: bash
-        run: |
-          export PATH=/c/lazarus/fpc/3.2.2/bin/x86_64-win64:$PATH
-
-          cd ./pascal-api-examples
-
-          pushd non-streaming-asr
-          ./run-zipformer-transducer.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-moonshine.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-whisper.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-nemo-transducer.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-nemo-ctc.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-sense-voice.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-telespeech-ctc.sh
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ./run-paraformer.sh
-
-          ./run-paraformer-itn.sh
-
-          rm -rf sherpa-onnx-*
-          echo "---"
-
-          ls -lh
-          popd
-
      - name:  Run Pascal test (Streaming ASR)
        shell: bash
        run: |
--- a/pascal-api-examples/non-streaming-asr/fire_red_asr.pas
+++ b/pascal-api-examples/non-streaming-asr/fire_red_asr.pas
@@ -0,0 +1,77 @@
+{ Copyright (c)  2025  Xiaomi Corporation }
+
+{
+This file shows how to use a non-streaming FireRedAsr AED model
+to decode files.
+
+You can download the model files from
+https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
+}
+
+program fire_red_asr;
+
+{$mode objfpc}
+
+uses
+  sherpa_onnx,
+  DateUtils,
+  SysUtils;
+
+var
+  Wave: TSherpaOnnxWave;
+  WaveFilename: AnsiString;
+
+  Config: TSherpaOnnxOfflineRecognizerConfig;
+  Recognizer: TSherpaOnnxOfflineRecognizer;
+  Stream: TSherpaOnnxOfflineStream;
+  RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
+
+  Start: TDateTime;
+  Stop: TDateTime;
+
+  Elapsed: Single;
+  Duration: Single;
+  RealTimeFactor: Single;
+begin
+  Initialize(Config);
+
+  Config.ModelConfig.FireRedAsr.Encoder := './sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/encoder.int8.onnx';
+  Config.ModelConfig.FireRedAsr.Decoder := './sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/decoder.int8.onnx';
+  Config.ModelConfig.Tokens := './sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/tokens.txt';
+  Config.ModelConfig.Provider := 'cpu';
+  Config.ModelConfig.NumThreads := 1;
+  Config.ModelConfig.Debug := False;
+
+  WaveFilename := './sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/test_wavs/0.wav';
+
+  Wave := SherpaOnnxReadWave(WaveFilename);
+
+  Recognizer := TSherpaOnnxOfflineRecognizer.Create(Config);
+  Stream := Recognizer.CreateStream();
+  Start := Now;
+
+  Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
+  Recognizer.Decode(Stream);
+
+  RecognitionResult := Recognizer.GetResult(Stream);
+
+  Stop := Now;
+
+  Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
+  Duration := Length(Wave.Samples) / Wave.SampleRate;
+  RealTimeFactor := Elapsed / Duration;
+
+  WriteLn(RecognitionResult.ToString);
+  WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
+  WriteLn(Format('Elapsed %.3f s', [Elapsed]));
+  WriteLn(Format('Wave duration %.3f s', [Duration]));
+  WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
+
+  {Free resources to avoid memory leak.
+
+  Note: You don't need to invoke them for this simple script.
+  However, you have to invoke them in your own large/complex project.
+  }
+  FreeAndNil(Stream);
+  FreeAndNil(Recognizer);
+end.
--- a/pascal-api-examples/non-streaming-asr/run-fire-red-asr.sh
+++ b/pascal-api-examples/non-streaming-asr/run-fire-red-asr.sh
@@ -0,0 +1,44 @@
+#!/usr/bin/env bash
+
+set -ex
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+SHERPA_ONNX_DIR=$(cd $SCRIPT_DIR/../.. && pwd)
+
+echo "SHERPA_ONNX_DIR: $SHERPA_ONNX_DIR"
+
+if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib  && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
+  mkdir -p ../../build
+  pushd ../../build
+  cmake \
+    -DCMAKE_INSTALL_PREFIX=./install \
+    -DSHERPA_ONNX_ENABLE_PYTHON=OFF \
+    -DSHERPA_ONNX_ENABLE_TESTS=OFF \
+    -DSHERPA_ONNX_ENABLE_CHECK=OFF \
+    -DBUILD_SHARED_LIBS=ON \
+    -DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
+    ..
+
+  cmake --build . --target install --config Release
+  ls -lh lib
+  popd
+fi
+
+if [ ! -f ./sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16.tar.bz2
+  tar xvf sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16.tar.bz2
+  rm sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16.tar.bz2
+  ls -lh sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16
+fi
+
+
+fpc \
+  -dSHERPA_ONNX_USE_SHARED_LIBS \
+  -Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
+  -Fl$SHERPA_ONNX_DIR/build/install/lib \
+  ./fire_red_asr.pas
+
+export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
+export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
+
+./fire_red_asr
--- a/sherpa-onnx/pascal-api/sherpa_onnx.pas
+++ b/sherpa-onnx/pascal-api/sherpa_onnx.pas
@@ -288,6 +288,12 @@ type
    function ToString: AnsiString;
  end;

+  TSherpaOnnxOfflineFireRedAsrModelConfig = record
+    Encoder: AnsiString;
+    Decoder: AnsiString;
+    function ToString: AnsiString;
+  end;
+
  TSherpaOnnxOfflineTdnnModelConfig = record
    Model: AnsiString;
    function ToString: AnsiString;
@@ -324,6 +330,7 @@ type
    TeleSpeechCtc: AnsiString;
    SenseVoice: TSherpaOnnxOfflineSenseVoiceModelConfig;
    Moonshine: TSherpaOnnxOfflineMoonshineModelConfig;
+    FireRedAsr: TSherpaOnnxOfflineFireRedAsrModelConfig;
    class operator Initialize({$IFDEF FPC}var{$ELSE}out{$ENDIF} Dest: TSherpaOnnxOfflineModelConfig);
    function ToString: AnsiString;
  end;
@@ -656,6 +663,10 @@ type
    Task: PAnsiChar;
    TailPaddings: cint32;
  end;
+  SherpaOnnxOfflineFireRedAsrModelConfig = record
+    Encoder: PAnsiChar;
+    Decoder: PAnsiChar;
+  end;
  SherpaOnnxOfflineMoonshineModelConfig = record
    Preprocessor: PAnsiChar;
    Encoder: PAnsiChar;
@@ -690,6 +701,7 @@ type
    TeleSpeechCtc: PAnsiChar;
    SenseVoice:  SherpaOnnxOfflineSenseVoiceModelConfig;
    Moonshine: SherpaOnnxOfflineMoonshineModelConfig;
+    FireRedAsr: SherpaOnnxOfflineFireRedAsrModelConfig;
  end;

  SherpaOnnxOfflineRecognizerConfig = record
@@ -1382,6 +1394,14 @@ begin
    [Self.Encoder, Self.Decoder, Self.Language, Self.Task, Self.TailPaddings]);
 end;

+function TSherpaOnnxOfflineFireRedAsrModelConfig.ToString: AnsiString;
+begin
+  Result := Format('TSherpaOnnxOfflineFireRedAsrModelConfig(' +
+    'Encoder := %s, ' +
+    'Decoder := %s)',
+    [Self.Encoder, Self.Decoder]);
+end;
+
 function TSherpaOnnxOfflineMoonshineModelConfig.ToString: AnsiString;
 begin
  Result := Format('TSherpaOnnxOfflineMoonshineModelConfig(' +
@@ -1434,13 +1454,15 @@ begin
    'BpeVocab := %s, ' +
    'TeleSpeechCtc := %s, ' +
    'SenseVoice := %s, ' +
-    'Moonshine := %s' +
+    'Moonshine := %s, ' +
+    'FireRedAsr := %s' +
    ')',
    [Self.Transducer.ToString, Self.Paraformer.ToString,
     Self.NeMoCtc.ToString, Self.Whisper.ToString, Self.Tdnn.ToString,
     Self.Tokens, Self.NumThreads, Self.Debug.ToString, Self.Provider,
     Self.ModelType, Self.ModelingUnit, Self.BpeVocab,
-     Self.TeleSpeechCtc, Self.SenseVoice.ToString, Self.Moonshine.ToString
+     Self.TeleSpeechCtc, Self.SenseVoice.ToString, Self.Moonshine.ToString,
+     Self.FireRedAsr.ToString
     ]);
 end;

@@ -1506,6 +1528,9 @@ begin
  C.ModelConfig.Moonshine.UncachedDecoder := PAnsiChar(Config.ModelConfig.Moonshine.UncachedDecoder);
  C.ModelConfig.Moonshine.CachedDecoder := PAnsiChar(Config.ModelConfig.Moonshine.CachedDecoder);

+  C.ModelConfig.FireRedAsr.Encoder := PAnsiChar(Config.ModelConfig.FireRedAsr.Encoder);
+  C.ModelConfig.FireRedAsr.Decoder := PAnsiChar(Config.ModelConfig.FireRedAsr.Decoder);
+
  C.LMConfig.Model := PAnsiChar(Config.LMConfig.Model);
  C.LMConfig.Scale := Config.LMConfig.Scale;