Audio
hyprwhspr
Native Linux dictation: fast, private, and hackable speech-to-text.
★ 1.2k starsMIT
View guide →// audio · DigitalPhonetics/IMS-Toucan
IMS-Toucan
IMS-Toucan is an open-source audio project by DigitalPhonetics (DigitalPhonetics/IMS-Toucan on GitHub). It has 2,208 GitHub stars and 318 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 25 January 2026.
ToucanTTS: fast, controllable text-to-speech for over 7,000 languages, from Stuttgart's IMS.
★ 2.2k stars⑂ 318 forksApache-2.0updated 2026-01
▦ visual — this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
so here's a placeholder. The guide below is complete regardless.
// project facts
Project facts
- Repository
- DigitalPhonetics/IMS-Toucan
- GitHub stars
- 2,208
- Forks
- 318
- License
- Apache-2.0
- Last updated on GitHub
- 25 January 2026
- Category
- Audio
- Listing
- Full install guide
// how it works
How it works
- Clarify your goal
Decide whether you need speech in one of the pretrained model's languages, or a fine-tuned model for your own voice or language data.
- Get the project
Clone the repository and install system packages such as libsndfile1, espeak-ng, and ffmpeg, then create a virtual environment and install the Python requirements.
- Install and configure
For custom work, prepare a training recipe by registering your dataset's audio paths and transcripts and copying a fine-tuning example.
- Run your first task
Run inference through the ToucanTTS interface or the provided reader scripts, or launch a training pipeline run for your recipe.
// features
Why people love it
Massively multilingual synthesis
Pretrained ToucanTTS models cover over 7,000 languages, with a published list of supported languages and a free hosted demo on Hugging Face.
Controllable output
Scaling parameters let you adjust duration, pitch-curve variance, and energy-curve variance at inference time.
Pretrained models on demand
Models download automatically when needed, and the interface supports setting language and speaker embeddings.
Fine-tuning recipes
Example pipelines for single-dataset and multilingual fine-tuning show how to adapt the model to new data with modest compute.
Full training pipelines
Train from scratch or fine-tune with the training pipeline script, including multi-GPU runs via torchrun and optional Weights and Biases logging.
Data quality tooling
A scorer utility can find and remove outlier samples with unusually high loss from a dataset cache before training.
Advertisement
// run it
Run it yourself
- Clone and install requirements
Install the listed system packages on Linux first, then set up the virtual environment.
git clone https://github.com/DigitalPhonetics/IMS-Toucan.git cd IMS-Toucan pip install --no-cache-dir -r requirements.txt - Run inference with the pretrained model
The text-to-file reader script synthesises a list of sentences into one audio file; pretrained weights download automatically.
python run_text_to_file_reader.py # see README inference section for interface options - Train or fine-tune a model
After preparing a recipe, launch the training pipeline; add --finetune to start from a checkpoint.
python run_training_pipeline.py --finetune
// use cases
What you can build
- Producing speech for a low-resource language that commercial TTS does not cover
- Fine-tuning a multilingual TTS voice on a small custom dataset
- Teaching or studying modern TTS architectures with a complete training codebase
- Generating narration audio on CPU-only machines using the pretrained model
Advertisement
// faq
Questions, answered
Who maintains it?
It is developed and maintained by researchers at the Institute for Natural Language Processing, University of Stuttgart, published under DigitalPhonetics.
Is it free to use?
Yes. The code and models are Apache-2.0 licensed and free, and the hosted Hugging Face demo is free to try.
Does it need a paid API key?
No API key is needed; models run locally in PyTorch and download from Hugging Face automatically.
How does it compare to a well-known alternative?
Commercial TTS APIs cover far fewer languages and charge per character; Toucan focuses on extreme multilingual coverage and trainability, at the cost of running and managing the model yourself.
How do I get started?
Clone the repo, install the system packages and requirements.txt, then run the text-to-file reader script; the pretrained model downloads on first use.
// keep exploring
Similar tools
Audio
Piper
Fast and local neural text-to-speech engine
★ 5.7k starsGPL-3.0
View guide →Audio
transcribe.cpp
Fast local speech-to-text across 16+ model families, powered by ggml.
★ 2.0k starsMIT
View guide →Audio
index-tts
IndexTTS: industrial-level, controllable zero-shot text-to-speech with voice cloning
★ 24.4k starsNOASSERTION
View guide →Audio
espeak-ng
eSpeak NG — the compact open-source speech synthesizer for 100+ languages, small enough to embed anywhere.
★ 6.9k starsGPL-3.0
View guide →Audio
marytts
MaryTTS — open-source, multilingual text-to-speech platform written in pure Java
★ 2.6k starsNOASSERTION
View guide →// alternatives
Alternatives in Audio
The most-starred other Audio projects in the LibreAI catalogue — the realistic shortlist if IMS-Toucan is not the right fit.
- MoneyPrinterTurbo★ 128.3k
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
- Whisper★ 109.9k
Robust Speech Recognition via Large-Scale Weak Supervision
- GPT-SoVITS★ 62.3k
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
- Real-Time-Voice-Cloning★ 60.2k
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
- whisper.cpp★ 54.1k
Port of OpenAI's Whisper model in C/C++
// more in audio
More in Audio
Nearby Audio projects in the catalogue, closest in stars to IMS-Toucan — keep exploring the shelf either side of it.
- hifi-gan★ 2.4k
GAN vocoder that turns mel-spectrograms into high-fidelity speech audio, fast enough for real-time use.
- WaveRNN★ 2.2k
WaveRNN vocoder plus Tacotron TTS in PyTorch: train your own neural speech synthesis on LJSpeech.
- VoiceMem★ 2.4k
VoiceMem — streaming dual-brain memory for voice agents: a left brain for facts, a right brain for emotion
- transcribe.cpp★ 2.0k
Fast local speech-to-text across 16+ model families, powered by ggml.
- pyttsx3★ 2.5k
Offline text-to-speech for Python — uses your operating system's own voices, no internet needed
- kokoro-tts★ 1.9k
Turn books and documents into natural speech with the Kokoro TTS model.
Get IMS-Toucan
Free and open — grab the source, star the repo, and start creating.