Piper En US Mira is a text-to-speech (TTS) voice model designed for the English language, notable for its ability to generate expressive, non-verbal vocalizations such as laughs, sighs, gasps, groans, and throat-clears. Unlike standard Piper voices that are limited to reading text, this model incorporates non-verbal expressions produced by the vocal model itself rather than relying on spliced-in sound effects.
05 kHz sampling rate. Its timbre is derived from the LJSpeech dataset. Piper En US Mira was fine-tuned in phoneme_ids mode, with certain unused phoneme IDs repurposed as marker tokens to trigger non-verbal sounds. It was trained on an emotive corpus distilled from ChatterBox Turbo (MIT), aligning these marker tokens with genuine non-verbal audio. By injecting a marker at synthesis time, users can prompt the model to perform the corresponding non-verbal expression during speech generation.
0 license. It is made available in the ONNX format and is intended for use within the Piper TTS system.
Piper En US Mira belongs to the class of expressive text-to-speech voice models and stands out for its integrated non-verbal synthesis capabilities, expanding the range of vocal expression available to developers and creators.
In the Voice, TTS & speech space, Piper En US Mira takes a focused approach. It enables developers to generate expressive English speech, including non-verbal sounds, using open-source TTS models. It is built as an open-source project for TTS developers and researchers. Piper En US Mira is open source under the GPL-3.0 license. The product ships for the web and the command line.
It is developed by BTimber, and the product first shipped in 2025. Development happens publicly on GitHub with 4.8k stars and 20 commits in the last 90 days. Key capabilities include Expressive TTS, non-verbal sounds, and english voice.
Latest indexed changes and source events
BTimber/piper-en_US-mira verified by the PulseGate indexer
Other apps tracked under the same category.
