Code-Inc/supertonic
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Supertonic 2

Browse Source
This commit is contained in:
ANLGBOY
2026-01-06 17:15:20 +09:00
parent ceedbbb835
commit e71c516714
44 changed files with 1105 additions and 581 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
rust/README.md
View File

@@ -4,6 +4,8 @@ This guide provides examples for running TTS inference using Rust.
## 📰 Update News
**2026.01.06** - 🎉 **Supertonic 2** released with multilingual support! Now supports English (`en`), Korean (`ko`), Spanish (`es`), Portuguese (`pt`), and French (`fr`). [Demo](https://huggingface.co/spaces/Supertone/supertonic-2) | [Models](https://huggingface.co/Supertone/supertonic-2)
**2025.12.10** - Added [6 new voice styles](https://huggingface.co/Supertone/supertonic/tree/b10dbaf18b316159be75b34d24f740008fddd381) (M3, M4, M5, F3, F4, F5). See [Voices](https://supertone-inc.github.io/supertonic-py/voices/) for details
**2025.12.08** - Optimized ONNX models via [OnnxSlim](https://github.com/inisis/OnnxSlim) now available on [Hugging Face Models](https://huggingface.co/Supertone/supertonic)
@@ -58,19 +60,21 @@ Process multiple voice styles and texts at once:
cargo run --release --bin example_onnx -- \
--batch \
--voice-style assets/voice_styles/M1.json,assets/voice_styles/F1.json \
--text "The sun sets behind the mountains, painting the sky in shades of pink and orange.|The weather is beautiful and sunny outside. A gentle breeze makes the air feel fresh and pleasant."
--text "The sun sets behind the mountains, painting the sky in shades of pink and orange.|오늘 아침에 공원을 산책했는데, 새소리와 바람 소리가 너무 기분 좋았어요." \
--lang en,ko
# Or using the binary directly
./target/release/example_onnx \
--batch \
--voice-style assets/voice_styles/M1.json,assets/voice_styles/F1.json \
--text "The sun sets behind the mountains, painting the sky in shades of pink and orange.|The weather is beautiful and sunny outside. A gentle breeze makes the air feel fresh and pleasant."
--text "The sun sets behind the mountains, painting the sky in shades of pink and orange.|오늘 아침에 공원을 산책했는데, 새소리와 바람 소리가 너무 기분 좋았어요." \
--lang en,ko
```
This will:
- Generate speech for 2 different voice-text pairs
- Use male voice (M1.json) for the first text
- Use female voice (F1.json) for the second text
- Generate speech for 2 different voice-text-language pairs
- Use male voice (M1.json) for the first text in English
- Use female voice (F1.json) for the second text in Korean
- Process both samples in a single batch
### Example 3: High Quality Inference
@@ -124,14 +128,16 @@ This will:
| `--onnx-dir` | str | `assets/onnx` | Path to ONNX model directory |
| `--total-step` | int | 5 | Number of denoising steps (higher = better quality, slower) |
| `--n-test` | int | 4 | Number of times to generate each sample |
| `--voice-style` | str+ | `assets/voice_styles/M1.json` | Voice style file path(s) |
| `--text` | str+ | (long default text) | Text(s) to synthesize |
| `--voice-style` | str+ | `assets/voice_styles/M1.json` | Voice style file path(s), comma-separated |
| `--text` | str+ | (long default text) | Text(s) to synthesize, pipe-separated |
| `--lang` | str+ | `en` | Language(s) for synthesis, comma-separated (en, ko, es, pt, fr) |
| `--save-dir` | str | `results` | Output directory |
| `--batch` | flag | False | Enable batch mode (multiple text-style pairs, disables automatic chunking) |
## Notes
- **Batch Processing**: When using `--batch`, the number of `--voice-style` files must match the number of `--text` entries
- **Multilingual Support**: Use `--lang` to specify the language for each text. Available: `en` (English), `ko` (Korean), `es` (Spanish), `pt` (Portuguese), `fr` (French)
- **Batch Processing**: When using `--batch`, the number of `--voice-style`, `--text`, and `--lang` entries must match
- **Automatic Chunking**: Without `--batch`, long texts are automatically split and concatenated with 0.3s pauses
- **Quality vs Speed**: Higher `--total-step` values produce better quality but take longer
- **GPU Support**: GPU mode is not supported yet

16
rust/src/example_onnx.rs
View File

@@ -42,6 +42,10 @@ struct Args {
#[arg(long, value_delimiter = '|', default_values_t = vec!["This morning, I took a walk in the park, and the sound of the birds and the breeze was so pleasant that I stopped for a long time just to listen.".to_string()])]
text: Vec<String>,
/// Language(s) for synthesis (en, ko, es, pt, fr)
#[arg(long, value_delimiter = ',', default_values_t = vec!["en".to_string()])]
lang: Vec<String>,
/// Output directory
#[arg(long, default_value = "results")]
save_dir: String,
@@ -61,6 +65,7 @@ fn main() -> Result<()> {
let n_test = args.n_test;
let voice_style_paths = &args.voice_style;
let text_list = &args.text;
let lang_list = &args.lang;
let save_dir = &args.save_dir;
let batch = args.batch;
@@ -72,6 +77,13 @@ fn main() -> Result<()> {
text_list.len()
);
}
if lang_list.len() != text_list.len() {
anyhow::bail!(
"Number of languages ({}) must match number of texts ({})",
lang_list.len(),
text_list.len()
);
}
}
let bsz = voice_style_paths.len();
@@ -90,11 +102,11 @@ fn main() -> Result<()> {
let (wav, duration) = if batch {
timer("Generating speech from text", || {
text_to_speech.batch(text_list, &style, total_step, speed)
text_to_speech.batch(text_list, lang_list, &style, total_step, speed)
})?
} else {
let (w, d) = timer("Generating speech from text", || {
text_to_speech.call(&text_list[0], &style, total_step, speed, 0.3)
text_to_speech.call(&text_list[0], &lang_list[0], &style, total_step, speed, 0.3)
})?;
(w, vec![d])
};

63
rust/src/helper.rs
View File

@@ -8,12 +8,19 @@ use serde_json;
use std::fs::File;
use std::io::BufReader;
use std::path::Path;
use anyhow::{Result, Context};
use anyhow::{Result, Context, bail};
use unicode_normalization::UnicodeNormalization;
use hound::{WavWriter, WavSpec, SampleFormat};
use rand_distr::{Distribution, Normal};
use regex::Regex;
// Available languages for multilingual TTS
pub const AVAILABLE_LANGS: &[&str] = &["en", "ko", "es", "pt", "fr"];
pub fn is_valid_lang(lang: &str) -> bool {
AVAILABLE_LANGS.contains(&lang)
}
// ============================================================================
// Configuration Structures
// ============================================================================
@@ -79,11 +86,11 @@ impl UnicodeProcessor {
Ok(UnicodeProcessor { indexer })
}
pub fn call(&self, text_list: &[String]) -> (Vec<Vec<i64>>, Array3<f32>) {
let processed_texts: Vec<String> = text_list
.iter()
.map(|t| preprocess_text(t))
.collect();
pub fn call(&self, text_list: &[String], lang_list: &[String]) -> Result<(Vec<Vec<i64>>, Array3<f32>)> {
let mut processed_texts: Vec<String> = Vec::new();
for (text, lang) in text_list.iter().zip(lang_list.iter()) {
processed_texts.push(preprocess_text(text, lang)?);
}
let text_ids_lengths: Vec<usize> = processed_texts
.iter()
@@ -108,16 +115,14 @@ impl UnicodeProcessor {
let text_mask = get_text_mask(&text_ids_lengths);
(text_ids, text_mask)
Ok((text_ids, text_mask))
}
}
pub fn preprocess_text(text: &str) -> String {
pub fn preprocess_text(text: &str, lang: &str) -> Result<String> {
// TODO: Need advanced normalizer for better performance
let mut text: String = text.nfkd().collect();
// FIXME: this should be fixed for non-English languages
// Remove emojis (wide Unicode range)
let emoji_pattern = Regex::new(r"[\x{1F600}-\x{1F64F}\x{1F300}-\x{1F5FF}\x{1F680}-\x{1F6FF}\x{1F700}-\x{1F77F}\x{1F780}-\x{1F7FF}\x{1F800}-\x{1F8FF}\x{1F900}-\x{1F9FF}\x{1FA00}-\x{1FA6F}\x{1FA70}-\x{1FAFF}\x{2600}-\x{26FF}\x{2700}-\x{27BF}\x{1F1E6}-\x{1F1FF}]+").unwrap();
text = emoji_pattern.replace_all(&text, "").to_string();
@@ -127,7 +132,6 @@ pub fn preprocess_text(text: &str) -> String {
("–", "-"), // en dash
("‑", "-"), // non-breaking hyphen
("—", "-"), // em dash
("¯", " "), // macron
("_", " "), // underscore
("\u{201C}", "\""), // left double quote
("\u{201D}", "\""), // right double quote
@@ -148,10 +152,6 @@ pub fn preprocess_text(text: &str) -> String {
text = text.replace(from, to);
}
// Remove combining diacritics // FIXME: this should be fixed for non-English languages
let diacritics_pattern = Regex::new(r"[\u{0302}\u{0303}\u{0304}\u{0305}\u{0306}\u{0307}\u{0308}\u{030A}\u{030B}\u{030C}\u{0327}\u{0328}\u{0329}\u{032A}\u{032B}\u{032C}\u{032D}\u{032E}\u{032F}]").unwrap();
text = diacritics_pattern.replace_all(&text, "").to_string();
// Remove special symbols
let special_symbols = ["♥", "☆", "♡", "©", "\\"];
for symbol in &special_symbols {
@@ -201,7 +201,15 @@ pub fn preprocess_text(text: &str) -> String {
}
}
text
// Validate language
if !is_valid_lang(lang) {
bail!("Invalid language: {}. Available: {:?}", lang, AVAILABLE_LANGS);
}
// Wrap text with language tags
text = format!("<{}>{}</{}>", lang, text, lang);
Ok(text)
}
pub fn text_to_unicode_values(text: &str) -> Vec<usize> {
@@ -505,15 +513,12 @@ where
}
pub fn sanitize_filename(text: &str, max_len: usize) -> String {
let text = if text.len() > max_len {
&text[..max_len]
} else {
text
};
// Take first max_len characters (Unicode code points, not bytes)
text.chars()
.take(max_len)
.map(|c| {
if c.is_ascii_alphanumeric() {
// is_alphanumeric() works with all Unicode letters and digits
if c.is_alphanumeric() {
c
} else {
'_'
@@ -570,6 +575,7 @@ impl TextToSpeech {
fn _infer(
&mut self,
text_list: &[String],
lang_list: &[String],
style: &Style,
total_step: usize,
speed: f32,
@@ -577,7 +583,7 @@ impl TextToSpeech {
let bsz = text_list.len();
// Process text
let (text_ids, text_mask) = self.text_processor.call(text_list);
let (text_ids, text_mask) = self.text_processor.call(text_list, lang_list)?;
let text_ids_array = {
let text_ids_shape = (bsz, text_ids[0].len());
@@ -676,18 +682,20 @@ impl TextToSpeech {
pub fn call(
&mut self,
text: &str,
lang: &str,
style: &Style,
total_step: usize,
speed: f32,
silence_duration: f32,
) -> Result<(Vec<f32>, f32)> {
let chunks = chunk_text(text, None);
let max_len = if lang == "ko" { 120 } else { 300 };
let chunks = chunk_text(text, Some(max_len));
let mut wav_cat: Vec<f32> = Vec::new();
let mut dur_cat: f32 = 0.0;
for (i, chunk) in chunks.iter().enumerate() {
let (wav, duration) = self._infer(&[chunk.clone()], style, total_step, speed)?;
let (wav, duration) = self._infer(&[chunk.clone()], &[lang.to_string()], style, total_step, speed)?;
let dur = duration[0];
let wav_len = (self.sample_rate as f32 * dur) as usize;
@@ -712,11 +720,12 @@ impl TextToSpeech {
pub fn batch(
&mut self,
text_list: &[String],
lang_list: &[String],
style: &Style,
total_step: usize,
speed: f32,
) -> Result<(Vec<f32>, Vec<f32>)> {
self._infer(text_list, style, total_step, speed)
self._infer(text_list, lang_list, style, total_step, speed)
}
}