Add scripts for exporting Piper TTS models to sherpa-onnx

.github/workflows/export-piper.yaml

@@ -0,0 +1,181 @@
+name: export-piper
+
+on:
+  push:
+    branches:
+      - export-piper
+  workflow_dispatch:
+
+concurrency:
+  group: export-piper-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-piper:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: ${{ matrix.index }}/${{ matrix.total }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest]
+        python-version: ["3.10"]
+        total: ["20"]
+        index: [
+          "0", "1", "2", "3", "4", "5", "6", "7", "8", "9",
+          "10", "11", "12", "13", "14", "15", "16", "17", "18", "19",
+           ]
+        # total: ["1"]
+        # index: ["0"]
+
+    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: |
+          python3 -m pip install --upgrade pip jinja2 iso639-lang onnx==1.17.0 onnxruntime==1.17.1 sherpa-onnx onnxmltools==1.13.0
+          python3 -m pip install "numpy<2" soundfile
+
+      - name: Generate script
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        shell: bash
+        run: |
+          cd scripts/piper
+
+          total=${{ matrix.total }}
+          index=${{ matrix.index }}
+
+          git config --global user.email "csukuangfj@gmail.com"
+          git config --global user.name "Fangjun Kuang"
+
+          git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/sherpa-onnx-tts-samples hf
+
+          python3 ./generate.py --total $total --index $index
+          chmod +x ./generate.sh
+          ls -lh
+
+      - name: Show script
+        shell: bash
+        run: |
+          cd scripts/piper
+          cat ./generate.sh
+
+      - name: Run script
+        shell: bash
+        run: |
+          cd scripts/piper
+          ./generate.sh
+
+      - name: Show generated mp3 files
+        shell: bash
+        run: |
+          cd scripts/piper
+          ls -lh hf/piper/mp3/*
+          echo "----"
+          ls -lh hf/piper/mp3/*/*
+
+      - name: Push generated mp3 files
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            cd scripts/piper/hf
+            git pull --rebase
+            git lfs track "*.mp3"
+            git status .
+            git add .
+            git commit -m 'Add mp3 files'
+            git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/sherpa-onnx-tts-samples main
+
+      - name: Show generated model files
+        shell: bash
+        run: |
+          cd scripts/piper
+          ls -lh *.tar.bz2
+
+      - name: Show generated model files(2)
+        shell: bash
+        run: |
+          cd scripts/piper
+          ls -lh release/
+
+      - 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 "csukuangfj@gmail.com"
+            git config --global user.name "Fangjun Kuang"
+
+            export GIT_LFS_SKIP_SMUDGE=1
+            export GIT_CLONE_PROTECTION_ACTIVE=false
+
+            dirs=(
+              vits-piper-de_DE-glados-high
+              vits-piper-de_DE-glados-low
+              vits-piper-de_DE-glados-medium
+              vits-piper-de_DE-glados_turret-high
+              vits-piper-de_DE-glados_turret-low
+              vits-piper-de_DE-glados_turret-medium
+              vits-piper-en_US-glados-high
+            )
+            for d in ${dirs[@]}; do
+              src=scripts/piper/release/$d
+              if [ ! -d $src ]; then
+                continue;
+              fi
+
+              rm -rf huggingface
+              git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d huggingface
+              cp -a $src/* ./huggingface
+              pushd huggingface
+              git lfs track "*.onnx"
+              git lfs track af_dict
+              git lfs track ar_dict
+              git lfs track cmn_dict
+              git lfs track da_dict en_dict fa_dict hu_dict ia_dict it_dict lb_dict phondata ru_dict ta_dict
+              git lfs track ur_dict yue_dict
+
+              git status
+              git add .
+              git status
+              git commit -m "add models"
+              git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d main
+              popd
+
+            done
+
+      - name: Release
+        if: github.repository_owner == 'csukuangfj'
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./scripts/piper/vits-piper-*.tar.bz2
+          overwrite: true
+          repo_name: k2-fsa/sherpa-onnx
+          repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
+          tag: tts-models
+
+      - name: Release
+        if: github.repository_owner == 'k2-fsa'
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./scripts/piper/vits-piper-*.tar.bz2
+          overwrite: true
+          tag: tts-models

scripts/piper/.gitignore

@@ -0,0 +1,5 @@
+*.sh
+*.onnx
+*.json
+MODEL_CARD
+generate_samples-vits-piper*.py

scripts/piper/add_meta_data.py

@@ -0,0 +1,117 @@
+#!/usr/bin/env python3
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+import argparse
+import json
+from typing import Any, Dict
+
+import onnx
+from iso639 import Lang
+
+
+def get_args():
+    # For en_GB-semaine-medium
+    # --name semaine
+    # --kind medium
+    # --lang en_GB
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+    parser.add_argument(
+        "--name",
+        type=str,
+        required=True,
+    )
+
+    parser.add_argument(
+        "--kind",
+        type=str,
+        required=True,
+    )
+
+    parser.add_argument(
+        "--lang",
+        type=str,
+        required=True,
+    )
+    return parser.parse_args()
+
+
+def add_meta_data(filename: str, meta_data: Dict[str, Any]):
+    """Add meta data to an ONNX model. It is changed in-place.
+
+    Args:
+      filename:
+        Filename of the ONNX model to be changed.
+      meta_data:
+        Key-value pairs.
+    """
+    model = onnx.load(filename)
+
+    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)
+
+    onnx.save(model, filename)
+
+
+def load_config(filename):
+    with open(filename, "r") as file:
+        config = json.load(file)
+    return config
+
+
+def generate_tokens(config):
+    id_map = config["phoneme_id_map"]
+    with open("tokens.txt", "w", encoding="utf-8") as f:
+        for s, i in id_map.items():
+            f.write(f"{s} {i[0]}\n")
+    print("Generated tokens.txt")
+
+
+# for en_US-lessac-medium.onnx
+# export LANG=en_US
+# export TYPE=lessac
+# export NAME=medium
+def main():
+    args = get_args()
+    print(args)
+    lang = args.lang
+
+    lang_iso = Lang(lang.split("_")[0])
+    print(lang, lang_iso)
+
+    kind = args.kind
+
+    name = args.name
+
+    # en_GB-alan-low.onnx.json
+    config = load_config(f"{lang}-{name}-{kind}.onnx.json")
+
+    print("generate tokens")
+    generate_tokens(config)
+
+    sample_rate = config["audio"]["sample_rate"]
+    if sample_rate == 22500:
+        print("Change sample rate from 22500 to 22050")
+        sample_rate = 22050
+
+    print("add model metadata")
+    meta_data = {
+        "model_type": "vits",
+        "comment": "piper",  # must be piper for models from piper
+        "language": lang_iso.name,
+        "voice": config["espeak"]["voice"],  # e.g., en-us
+        "has_espeak": 1,
+        "n_speakers": config["num_speakers"],
+        "sample_rate": sample_rate,
+    }
+    print(meta_data)
+    add_meta_data(f"{lang}-{name}-{kind}.onnx", meta_data)
+
+
+main()

scripts/piper/dynamic_quantization.py

@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+import argparse
+
+import onnxmltools
+from onnxmltools.utils.float16_converter import convert_float_to_float16
+from onnxruntime.quantization import QuantType, quantize_dynamic
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--input",
+        type=str,
+        required=True,
+    )
+    parser.add_argument(
+        "--output-fp16",
+        type=str,
+        required=True,
+    )
+
+    parser.add_argument(
+        "--output-int8",
+        type=str,
+        required=True,
+    )
+    return parser.parse_args()
+
+
+# for op_block_list, see also
+# https://github.com/microsoft/onnxruntime/blob/089c52e4522491312e6839af146a276f2351972e/onnxruntime/python/tools/transformers/float16.py#L115
+#
+# libc++abi: terminating with uncaught exception of type Ort::Exception:
+# Type Error: Type (tensor(float16)) of output arg (/dp/RandomNormalLike_output_0)
+# of node (/dp/RandomNormalLike) does not match expected type (tensor(float)).
+#
+# libc++abi: terminating with uncaught exception of type Ort::Exception:
+# This is an invalid model. Type Error: Type 'tensor(float16)' of input
+# parameter (/enc_p/encoder/attn_layers.0/Constant_84_output_0) of
+# operator (Range) in node (/Range_1) is invalid.
+def export_onnx_fp16(onnx_fp32_path, onnx_fp16_path):
+    onnx_fp32_model = onnxmltools.utils.load_model(onnx_fp32_path)
+    onnx_fp16_model = convert_float_to_float16(
+        onnx_fp32_model,
+        keep_io_types=True,
+        op_block_list=[
+            "RandomNormalLike",
+            "Range",
+        ],
+    )
+    onnxmltools.utils.save_model(onnx_fp16_model, onnx_fp16_path)
+
+
+def main():
+    args = get_args()
+    print(args)
+
+    in_filename = args.input
+    output_fp16 = args.output_fp16
+    output_int8 = args.output_int8
+
+    quantize_dynamic(
+        model_input=in_filename,
+        model_output=output_int8,
+        weight_type=QuantType.QUInt8,
+    )
+
+    export_onnx_fp16(in_filename, output_fp16)
+
+
+if __name__ == "__main__":
+    main()