Moshi: a speech-text foundation model for real-time dialogue
About
We introduce Moshi, a speech-text foundation model and full-duplex spoken dialogue framework. Current systems for spoken dialogue rely on pipelines of independent components, namely voice activity detection, speech recognition, textual dialogue and text-to-speech. Such frameworks cannot emulate the experience of real conversations. First, their complexity induces a latency of several seconds between interactions. Second, text being the intermediate modality for dialogue, non-linguistic information that modifies meaning -- such as emotion or non-speech sounds -- is lost in the interaction. Finally, they rely on a segmentation into speaker turns, which does not take into account overlapping speech, interruptions and interjections. Moshi solves these independent issues altogether by casting spoken dialogue as speech-to-speech generation. Starting from a text language model backbone, Moshi generates speech as tokens from the residual quantizer of a neural audio codec, while modeling separately its own speech and that of the user into parallel streams. This allows for the removal of explicit speaker turns, and the modeling of arbitrary conversational dynamics. We moreover extend the hierarchical semantic-to-acoustic token generation of previous work to first predict time-aligned text tokens as a prefix to audio tokens. Not only this "Inner Monologue" method significantly improves the linguistic quality of generated speech, but we also illustrate how it can provide streaming speech recognition and text-to-speech. Our resulting model is the first real-time full-duplex spoken large language model, with a theoretical latency of 160ms, 200ms in practice, and is available at https://github.com/kyutai-labs/moshi.
Key insights
What problem does Moshi solve?
Moshi removes the latency, text bottleneck, and rigid turn-taking of cascaded voice assistants by modeling spoken dialogue as direct speech-to-speech generation with parallel user and system audio streams.
How does Moshi differ from prior work?
Unlike dGSLM or Spectron, Moshi jointly models semantic and acoustic tokens in a streaming, multi-stream architecture. It achieves 160ms theoretical latency and reports state-of-the-art spoken question answering accuracy.
What are the practical uses?
Moshi enables real-time full-duplex voice conversations with overlapping speech. A public web demo exists, but the paper documents no production deployment beyond this prototype benchmark.
What limitations do the authors note?
The authors note Moshi underperforms its text backbone on complex TriviaQA. They also found signal-based audio watermarking fails under codec compression, and that quantization below 4 bits degrades linguistic quality.