kokorosherpaonnx

Kokoro Sherpa package with em_santa

This directory contains a Sherpa-ONNX Kokoro v1.0 package extended with the Spanish voice em_santa.

Files

• model.onnx — the official Kokoro v1.0 model with n_speakers changed from 53 to 54 and metadata for speaker ID 53.
• voices.bin — the official 53 voice embeddings plus em_santa appended as speaker ID 53.
• kokoro-multi-lang-v1_0-em-santa.tar.bz2 — complete ready-to-extract package.
• manifest.json — source information and SHA-256 checksums.

The other files required by Kokoro (tokens.txt, lexicons, rule FSTs and espeak-ng-data) must remain alongside these files from the original Sherpa package. Do not use only voices.bin: Sherpa validates the voice count in model.onnx, so both patched files are required.

Sherpa-ONNX configuration

Configure the Kokoro model with the patched files and select speaker ID 53:

model   = /path/to/model.onnx
voices  = /path/to/voices.bin
tokens  = /path/to/tokens.txt
lexicon = /path/to/lexicon-us-en.txt,/path/to/lexicon-zh.txt
data_dir = /path/to/espeak-ng-data
lang    = es
sid     = 53

In the Sherpa-ONNX native API this corresponds to an OfflineTtsKokoroModelConfig whose model, voices, tokens, lexicon and dataDir point to the files above. Set OfflineTtsConfig.max_num_sentences to the value appropriate for the host application and pass sid=53 in the generation configuration. The voice is selected per generation; it is not installed as a system TTS voice.

Important

Keep the original Sherpa-ONNX and Kokoro license/attribution files when redistributing this package. This generated package does not add or replace any upstream license. Sources are listed in manifest.json.

Generated with tools/build_kokoro_sherpa_voice_package.py in the BookReader repository.

Detailed integration instructions

The complete model directory must contain the official Kokoro support files as well as the two replacement files in this folder. Replace the official model.onnx and voices.bin; keep tokens.txt, all lexicons, all rule FST files, and the complete espeak-ng-data directory.

For Spanish Santa, configure the Kokoro model with lang = "es" and generate with sid = 53. Do not use the voice name as sid and do not use only voices.bin: Sherpa validates the embedding count against the n_speakers metadata in model.onnx.

Native configuration:

config.model.kokoro.model = "/path/to/model.onnx";
config.model.kokoro.voices = "/path/to/voices.bin";
config.model.kokoro.tokens = "/path/to/tokens.txt";
config.model.kokoro.lexicon =
    "/path/to/lexicon-us-en.txt,/path/to/lexicon-zh.txt";
config.model.kokoro.data_dir = "/path/to/espeak-ng-data";
config.model.kokoro.lang = "es";
generation.sid = 53;
generation.speed = 1.0f;

Android Kotlin uses the same paths and IDs:

val kokoro = OfflineTtsKokoroModelConfig(
    model = File(modelDir, "model.onnx").absolutePath,
    voices = File(modelDir, "voices.bin").absolutePath,
    tokens = File(modelDir, "tokens.txt").absolutePath,
    lexicon = listOf(
        File(modelDir, "lexicon-us-en.txt").absolutePath,
        File(modelDir, "lexicon-zh.txt").absolutePath,
    ).joinToString(","),
    dataDir = File(modelDir, "espeak-ng-data").absolutePath,
    lang = "es",
)
val audio = tts.generateWithConfig(
    "Texto en español.",
    GenerationConfig(sid = 53, speed = 1.0f),
)

To verify an installation, check that model.onnx reports n_speakers=54, speaker ID 53 is em_santa, and voices.bin is 28,200,960 bytes. The final voice block is 522,240 bytes and uses float32 values with shape 510 x 1 x 256. Keep the original Sherpa-ONNX and Kokoro license and attribution files when redistributing the package.

Model card and licensing

This repository is a derived distribution of the Sherpa-ONNX Kokoro v1.0 package. The original package is distributed under the Apache License 2.0; the complete license text is included as LICENSE in the original package. The em_santa embedding is sourced from the public ONNX Community Kokoro v1.0 repository and must retain the applicable upstream terms.

Upstream references:

• Sherpa-ONNX: https://github.com/k2-fsa/sherpa-onnx
• Official Sherpa Kokoro package: https://huggingface.co/csukuangfj/kokoro-multi-lang-v1_0
• Kokoro voice catalogue and license: https://huggingface.co/hexgrad/Kokoro-82M
• em_santa source embedding: https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX

This repository does not claim ownership of the Kokoro model or its voice embeddings. Redistributors must preserve this notice, the included LICENSE file, and all applicable notices from Sherpa-ONNX, Kokoro, espeak-ng and other bundled dependencies.