TTS

README.md

@@ -1,6 +1,6 @@
 # STT Runner
 
-Speech-to-Text transcription using sherpa-onnx + Qwen3-ASR.
+Speech-to-Text transcription using sherpa-onnx + Qwen3-ASR, plus Text-to-Speech with ZipVoice (zero-shot voice cloning).
 
 ## Installation
 
@@ -11,16 +11,27 @@ python3 -m venv .venv
 
 ## Usage
 
+### Speech-to-Text
+
 ```bash
 python stt_runner.py [--language=Indonesian] audio1.wav audio2.wav ...
 ```
 
+### Text-to-Speech
+
+```bash
+python tts_runner.py [--output=out.wav] [--ref-audio=ref.wav] [--ref-text="..."] "text to speak"
+```
+
+Output defaults to `output.wav`. The reference audio/text (voice to clone) is set in `config/tts.py` and can be overridden per-run with `--ref-audio` / `--ref-text` (the text must match the audio exactly).
+
 ## Configuration
 
 Model paths and inference parameters are hardcoded in `config/`:
 
 - `config/model.py` — model paths (conv_frontend, encoder, decoder, tokenizer under `models/`)
 - `config/asr.py` — inference params: `LANGUAGE`, `HOTWORDS`, `NUM_THREADS`, `PROVIDER`, `SAMPLE_RATE`, `FEATURE_DIM`, `MAX_TOTAL_LEN`, `MAX_NEW_TOKENS`
+- `config/tts.py` — TTS model paths, `REFERENCE_AUDIO`, `REFERENCE_TEXT`, `OUTPUT_FILE`, `NUM_THREADS`, `PROVIDER`, `NUM_STEPS`
 
 `LANGUAGE` defaults to `""` (all languages / auto-detect). Passing `--language` on the CLI overrides it.
 
@@ -35,4 +46,14 @@ wget -O models/model_1.7B/decoder.int8.onnx  "$BASE/model_1.7B/decoder.int8.onnx
 for f in vocab.json merges.txt tokenizer_config.json preprocessor_config.json config.json chat_template.json; do
   wget -O "models/tokenizer/$f" "$BASE/tokenizer/$f"
 done
+```
+
+## Download Model (ZipVoice TTS)
+
+```bash
+mkdir -p models/zipvoice
+wget -qO- https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-zipvoice-distill-int8-zh-en-emilia.tar.bz2 \
+  | tar xjf - -C models/zipvoice --strip-components=1
+wget -O models/zipvoice/vocos_24khz.onnx \
+  https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/vocos_24khz.onnx
 ```

config/tts.py

@@ -0,0 +1,23 @@
+from config.model import MODEL_DIR
+
+ZIPVOICE = MODEL_DIR / "zipvoice"
+
+TOKENS   = ZIPVOICE / "tokens.txt"
+ENCODER  = ZIPVOICE / "encoder.int8.onnx"
+DECODER  = ZIPVOICE / "decoder.int8.onnx"
+DATA_DIR = ZIPVOICE / "espeak-ng-data"
+LEXICON  = ZIPVOICE / "lexicon.txt"
+VOCODER  = ZIPVOICE / "vocos_24khz.onnx"
+
+# Reference voice for zero-shot voice cloning.
+# REFERENCE_TEXT must match what is spoken in REFERENCE_AUDIO exactly,
+# otherwise the cloned voice quality will noticeably degrade.
+REFERENCE_AUDIO = ZIPVOICE / "test_wavs" / "leijun-1.wav"
+REFERENCE_TEXT  = "那还是三十六年前, 一九八七年. 我呢考上了武汉大学的计算机系."
+
+NUM_THREADS = 2
+PROVIDER    = "cpu"
+NUM_STEPS   = 4 # Generation quality/speed tradeoff (higher = better, slower)
+
+# Default output filename.
+OUTPUT_FILE = "output.wav"

core/tts_room.py

@@ -0,0 +1,7 @@
+import argparse
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--output",      type=str, default=None, help="Output WAV file, overrides config OUTPUT_FILE")
+parser.add_argument("--ref-audio",   type=str, default=None, help="Reference audio for voice cloning, overrides config REFERENCE_AUDIO")
+parser.add_argument("--ref-text",    type=str, default=None, help="Reference text matching --ref-audio, overrides config REFERENCE_TEXT")
+parser.add_argument("text", type=str, help="Text to synthesize into speech")

tts_runner.py

@@ -0,0 +1,75 @@
+import sys
+import time
+from pathlib import Path
+import sherpa_onnx, soundfile as sf
+from core import tts_room as ap
+from config.tts import (
+    TOKENS, ENCODER, DECODER, DATA_DIR, LEXICON, VOCODER,
+    REFERENCE_AUDIO, REFERENCE_TEXT, NUM_THREADS, PROVIDER, NUM_STEPS, OUTPUT_FILE,
+)
+
+def tts_run(args):
+
+    ref_audio = Path(args.ref_audio) if args.ref_audio is not None else Path(REFERENCE_AUDIO)
+    ref_text  = args.ref_text       if args.ref_text  is not None else REFERENCE_TEXT
+
+    if not ref_audio.is_file():
+        print(f"Reference audio not found: {ref_audio}", file=sys.stderr)
+        return
+
+    print("Loading model...")
+    tts_config = sherpa_onnx.OfflineTtsConfig(
+        model=sherpa_onnx.OfflineTtsModelConfig(
+            zipvoice=sherpa_onnx.OfflineTtsZipvoiceModelConfig(
+                tokens   = str(TOKENS),
+                encoder  = str(ENCODER),
+                decoder  = str(DECODER),
+                data_dir = str(DATA_DIR),
+                lexicon  = str(LEXICON),
+                vocoder  = str(VOCODER),
+            ),
+            debug       = False,
+            num_threads = NUM_THREADS,
+            provider    = PROVIDER,
+        )
+    )
+    if not tts_config.validate():
+        raise ValueError("Invalid TTS config. Please read the previous error messages.")
+
+    tts = sherpa_onnx.OfflineTts(tts_config)
+    print("Model ready!")
+
+    reference_audio, sr = sf.read(str(ref_audio), dtype="float32")
+    if reference_audio.ndim > 1:
+        reference_audio = reference_audio[:, 0]
+
+    gen_config = sherpa_onnx.GenerationConfig()
+    gen_config.reference_audio       = reference_audio
+    gen_config.reference_sample_rate = sr
+    gen_config.reference_text        = ref_text
+    gen_config.num_steps             = NUM_STEPS
+
+    output_file = args.output if args.output is not None else OUTPUT_FILE
+
+    print("Generating...")
+    start = time.time()
+    audio = tts.generate(args.text, gen_config)
+    elapsed = time.time() - start
+
+    if len(audio.samples) == 0:
+        print("Error in generating audio. Please read the previous error messages.", file=sys.stderr)
+        return
+
+    sf.write(output_file, audio.samples, samplerate=audio.sample_rate, subtype="PCM_16")
+
+    duration = len(audio.samples) / audio.sample_rate
+    print()
+    print(f"Output       : {output_file}")
+    print(f"Text         : {args.text}")
+    print(f"Duration     : {duration:.3f} s")
+    print(f"Elapsed      : {elapsed:.3f} s")
+    print(f"RTF          : {elapsed:.3f} / {duration:.3f} = {elapsed / duration:.3f}")
+    print()
+
+if __name__ == "__main__":
+    tts_run( ap.parser.parse_args() )

usage-tts.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PYTHON="${PYTHON:-$BASE_DIR/.venv/bin/python}"
+
+if [ ! -x "$PYTHON" ]; then
+    echo "Venv not found. Create it with: python3 -m venv .venv && .venv/bin/pip install -r requirements.txt" >&2
+    exit 1
+fi
+
+if [ $# -lt 1 ]; then
+    echo "Usage: $0 <text> [--output <file.wav>] [--ref-audio <ref.wav>] [--ref-text <text>]" >&2
+    echo "       Synthesize speech with ZipVoice zero-shot TTS (sherpa-onnx)." >&2
+    echo "       Reference audio/text defaults are set in config/tts.py." >&2
+    exit 1
+fi
+
+"$PYTHON" "$BASE_DIR/tts_runner.py" "$@"

usage.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PYTHON="${PYTHON:-$BASE_DIR/.venv/bin/python}"
+
+if [ ! -x "$PYTHON" ]; then
+    echo "Venv not found. Create it with: python3 -m venv .venv && .venv/bin/pip install -r requirements.txt" >&2
+    exit 1
+fi
+
+if [ $# -lt 1 ]; then
+    echo "Usage: $0 <audio_file...>" >&2
+    echo "       Transcribe audio with Qwen3-ASR 1.7B (sherpa-onnx)." >&2
+    exit 1
+fi
+
+"$PYTHON" "$BASE_DIR/stt_runner.py" "$@"