Add Supertonic2 TTS support (#3094)

This pull request significantly enhances the offline Text-to-Speech capabilities of sherpa-onnx by integrating the Supertonic 2 TTS system. It introduces a complete pipeline for Supertonic 2, encompassing model configuration, a sophisticated multilingual text preprocessing module, and the core inference logic. The changes enable users to perform high-quality, on-device speech synthesis with support for multiple languages and advanced features like batch processing and customizable synthesis parameters, thereby broadening the application scope and improving the overall performance of the TTS module.

Author: Wasser1462

.github/workflows/export-supertonic.yaml

@@ -0,0 +1,110 @@
+name: export-supertonic-to-int8-onnx
+
+on:
+  push:
+    branches:
+      - supertonic
+
+  workflow_dispatch:
+
+concurrency:
+  group: export-supertonic-to-int8-onnx-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-supertonic-to-onnx:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: export supertonic int8
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Install Python dependencies
+        shell: bash
+        run: |
+          pip install numpy onnx onnxruntime modelscope
+
+      - name: Run
+        shell: bash
+        run: |
+          cd scripts/supertonic
+          ./run.sh
+
+      - name: Collect results
+        shell: bash
+        run: |
+          src=scripts/supertonic
+          d=supertonic-int8
+
+          mkdir $d
+          cp -a LICENSE $d/LICENSE
+          cp -a $src/README.md $d/README.md
+          cp -v $src/onnx_int8/*.int8.onnx $d/
+          [ -f $src/assets/onnx/unicode_indexer.bin ] && cp -v $src/assets/onnx/unicode_indexer.bin $d/
+          [ -f $src/assets/onnx/tts.json ] && cp -v $src/assets/onnx/tts.json $d/
+          [ -f $src/assets/voice_styles/voice.bin ] && cp -v $src/assets/voice_styles/voice.bin $d/voice.bin
+          ls -lh $d/
+          tar cjfv $d.tar.bz2 $d
+
+          ls -lh $d.tar.bz2
+
+      - name: Release
+        if: github.repository_owner == 'csukuangfj'
+        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
+
+      - name: Release
+        if: github.repository_owner == 'k2-fsa'
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./*.tar.bz2
+          overwrite: true
+          tag: tts-models
+
+      - 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
+
+            d=supertonic-int8
+            if [[ ! -d $d ]]; then
+              echo "$d does not exist"
+              exit 0
+            fi
+
+            rm -rf huggingface
+            git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d huggingface
+            cd huggingface
+            rm -rf ./*
+
+            git lfs track "*.onnx"
+            cp -a ../$d/* ./
+
+            git add .
+            ls -lh
+            git status
+            git commit -m "add supertonic int8 models" || true
+            git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d main || true

scripts/supertonic/README.md

@@ -0,0 +1,20 @@
+# Supertonic TTS INT8 Quantization
+
+Quantize [Supertonic](https://github.com/supertone-inc/supertonic) TTS ONNX models to INT8 for on-device deployment.
+
+## Overview
+
+- **Pipeline**: `gen_calib_configs` → `dump_inputs` → `convert`; stage 4 generates **.bin** assets when JSONs exist: `generate_voices_bin.py`, `generate_indexer_bin.py`. Runtime loads **tts.json** for TTS config.
+- **Quantization**: duration_predictor, text_encoder, vector_estimator → dynamic INT8; vocoder → static INT8 (calibration from dumped data).
+- **Voice**: Runtime loads one **`voice.bin`**. Generate with `python3 generate_voices_bin.py [input_dir] [output_bin]`. Pass `--supertonic-voice-style=/path/to/voice.bin`. Use `--sid` 0..N-1 to select speaker.
+- **Unicode indexer**: Runtime uses **`unicode_indexer.bin`**. Generate with `python3 generate_indexer_bin.py [json_path] [bin_path]`. Pass `--supertonic-unicode-indexer=/path/to/unicode_indexer.bin`.
+- **TTS config**: Runtime loads **`tts.json`**. Pass `--supertonic-tts-json=/path/to/tts.json`.
+
+## Usage
+
+```bash
+./run.sh              # Run all stages (0–4)
+./run.sh 4            # Only generate voice.bin, unicode_indexer.bin
+```
+
+**Stages:** 0 = download models, 1 = gen calib configs, 2 = dump calib data, 3 = quantize, 4 = generate `voice.bin`, `unicode_indexer.bin`.