Export kokoro to sherpa-onnx

.github/workflows/export-kokoro.yaml

@@ -0,0 +1,122 @@
+name: export-kokoro-to-onnx
+
+on:
+  push:
+    branches:
+      - export-kokoro
+
+  workflow_dispatch:
+
+concurrency:
+  group: export-kokoro-to-onnx-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-kokoro-to-onnx:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: export kokoro
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest]
+        python-version: ["3.10"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install Python dependencies
+        shell: bash
+        run: |
+          pip install -q "numpy<=1.26.4" onnx==1.16.0 onnxruntime==1.17.1 librosa soundfile piper_phonemize -f https://k2-fsa.github.io/icefall/piper_phonemize.html
+
+      - name: Run
+        shell: bash
+        run: |
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/espeak-ng-data.tar.bz2
+          tar xf espeak-ng-data.tar.bz2
+          rm espeak-ng-data.tar.bz2
+          cd scripts/kokoro
+          ./run.sh
+
+      - name: Collect results
+        shell: bash
+        run: |
+          src=scripts/kokoro
+
+          d=kokoro-en-v0_19
+          mkdir $d
+          cp -a LICENSE $d/LICENSE
+          cp -a espeak-ng-data $d/
+          cp -v $src/kokoro-v0_19_hf.onnx $d/model.onnx
+          cp -v $src/voices.bin $d/
+          cp -v $src/tokens.txt $d/
+          cp -v $src/README-new.md $d/README.md
+          ls -lh $d/
+          tar cjfv $d.tar.bz2 $d
+          rm -rf $d
+
+          ls -h $.tar.bz2
+
+      - name: Publish to huggingface
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            git config --global user.email "[email protected]"
+            git config --global user.name "Fangjun Kuang"
+
+            rm -rf huggingface
+            export GIT_LFS_SKIP_SMUDGE=1
+            export GIT_CLONE_PROTECTION_ACTIVE=false
+
+            git clone https://csukuangfj:[email protected]/csukuangfj/kokoro-en-v0_19 huggingface
+            cd huggingface
+            rm -rf ./*
+            git fetch
+            git pull
+
+            git lfs track "cmn_dict"
+            git lfs track "ru_dict"
+            git lfs track "*.wav"
+
+            cp -a ../espeak-ng-data ./
+            mkdir -p test_wavs
+
+            cp -v ../scripts/kokoro/kokoro-v0_19_hf.onnx ./model.onnx
+
+            cp -v ../scripts/kokoro/kokoro-v0_19_hf-*.wav ./test_wavs/
+
+            cp -v ../scripts/kokoro/tokens.txt .
+            cp -v ../scripts/kokoro/voices.bin .
+            cp -v ../scripts/kokoro/README-new.md ./README.md
+            cp -v ../LICENSE ./
+
+            git lfs track "*.onnx"
+            git add .
+
+            ls -lh
+
+            git status
+
+            git commit -m "add models"
+            git push https://csukuangfj:[email protected]/csukuangfj/kokoro-en-v0_19 main || true
+
+      - name: Release
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./*.tar.bz2
+          overwrite: true
+          repo_name: k2-fsa/sherpa-onnx
+          repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
+          tag: tts-models

scripts/kokoro/.gitignore

@@ -0,0 +1,3 @@
+voices.json
+voices.bin
+README-new.md

scripts/kokoro/README.md

@@ -0,0 +1,10 @@
+# Introduction
+
+This folder contains scripts for adding meta data to models
+from https://github.com/thewh1teagle/kokoro-onnx/releases/tag/model-files
+
+See also
+https://huggingface.co/hexgrad/Kokoro-82M/tree/main
+and
+https://huggingface.co/spaces/hexgrad/Kokoro-TTS
+

scripts/kokoro/add-meta-data.py

@@ -0,0 +1,107 @@
+#!/usr/bin/env python3
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+
+import argparse
+import json
+from pathlib import Path
+
+import numpy as np
+import onnx
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", type=str, required=True, help="input and output onnx model"
+    )
+
+    parser.add_argument("--voices", type=str, required=True, help="Path to voices.json")
+    return parser.parse_args()
+
+
+def load_voices(filename):
+    with open(filename) as f:
+        voices = json.load(f)
+    for key in voices:
+        voices[key] = np.array(voices[key], dtype=np.float32)
+    return voices
+
+
+def get_vocab():
+    _pad = "$"
+    _punctuation = ';:,.!?¡¿—…"«»“” '
+    _letters = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
+    _letters_ipa = "ɑɐɒæɓʙβɔɕçɗɖðʤəɘɚɛɜɝɞɟʄɡɠɢʛɦɧħɥʜɨɪʝɭɬɫɮʟɱɯɰŋɳɲɴøɵɸθœɶʘɹɺɾɻʀʁɽʂʃʈʧʉʊʋⱱʌɣɤʍχʎʏʑʐʒʔʡʕʢǀǁǂǃˈˌːˑʼʴʰʱʲʷˠˤ˞↓↑→↗↘'̩'ᵻ"
+    symbols = [_pad] + list(_punctuation) + list(_letters) + list(_letters_ipa)
+    dicts = {}
+    for i in range(len((symbols))):
+        dicts[symbols[i]] = i
+    return dicts
+
+
+def generate_tokens():
+    token2id = get_vocab()
+    with open("tokens.txt", "w", encoding="utf-8") as f:
+        for s, i in token2id.items():
+            f.write(f"{s} {i}\n")
+
+
+def main():
+    args = get_args()
+    print(args.model, args.voices)
+
+    model = onnx.load(args.model)
+    voices = load_voices(args.voices)
+
+    if Path("./tokens.txt").is_file():
+        print("./tokens.txt exist, skip generating it")
+    else:
+        generate_tokens()
+
+    keys = list(voices.keys())
+    print(",".join(keys))
+
+    if Path("./voices.bin").is_file():
+        print("./voices.bin exists, skip generating it")
+    else:
+        with open("voices.bin", "wb") as f:
+            for k in keys:
+                f.write(voices[k].tobytes())
+
+    meta_data = {
+        "model_type": "kokoro",
+        "language": "English",
+        "has_espeak": 1,
+        "sample_rate": 24000,
+        "version": 1,
+        "voice": "en-us",
+        "style_dim": ",".join(map(str, voices[keys[0]].shape)),
+        "n_speakers": len(keys),
+        "speaker_names": ",".join(keys),
+        "model_url": "https://github.com/thewh1teagle/kokoro-onnx/releases/tag/model-files",
+        "see_also": "https://huggingface.co/spaces/hexgrad/Kokoro-TTS",
+        "see_also_2": "https://huggingface.co/hexgrad/Kokoro-82M",
+        "maintainer": "k2-fsa",
+    }
+
+    print(model.metadata_props)
+
+    while len(model.metadata_props):
+        model.metadata_props.pop()
+
+    for key, value in meta_data.items():
+        meta = model.metadata_props.add()
+        meta.key = key
+        meta.value = str(value)
+    print("--------------------")
+
+    print(model.metadata_props)
+
+    onnx.save(model, args.model)
+
+    print(f"Please see {args.model}, ./voices.bin, and ./tokens.txt")
+
+
+if __name__ == "__main__":
+    main()

scripts/kokoro/run.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+set -ex
+
+cat > README-new.md <<EOF
+# Introduction
+
+Files in this folder are from
+https://github.com/thewh1teagle/kokoro-onnx/releases/tag/model-files
+
+Please see also
+https://huggingface.co/hexgrad/Kokoro-82M
+and
+https://huggingface.co/hexgrad/Kokoro-82M/discussions/14
+EOF
+
+files=(
+kokoro-v0_19_hf.onnx
+# kokoro-v0_19.onnx
+# kokoro-quant.onnx
+# kokoro-quant-convinteger.onnx
+voices.json
+)
+
+for f in ${files[@]}; do
+  if [ ! -f ./$f ]; then
+    curl -SL -O https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files/$f
+  fi
+done
+
+models=(
+# kokoro-v0_19
+# kokoro-quant
+# kokoro-quant-convinteger
+kokoro-v0_19_hf
+)
+
+for m in ${models[@]}; do
+  ./add-meta-data.py --model $m.onnx --voices ./voices.json
+done
+
+ls -l
+echo "----------"
+ls -lh
+
+for m in ${models[@]}; do
+  ./test.py --model $m.onnx --voices-bin ./voices.bin --tokens ./tokens.txt
+done
+ls -lh