Nari Labs Ships Qwen3-Based TTS and ASR Models
The startup pairs speech synthesis and recognition built on Qwen3, pitching accuracy, low latency, and lower cost.
Nari Labs has released a pair of speech models built on Qwen3: a text-to-speech (TTS) system and an automatic speech recognition (ASR) system. The company frames the launch around three practical goals that matter to anyone building voice applications — accuracy, low latency, and cost — rather than raw novelty.
According to Nari Labs, the models lead on Coval voice AI benchmarks, a suite that evaluates conversational voice performance. By offering both synthesis and recognition, Nari is targeting the full round-trip of a voice agent: understanding what a user says and responding in natural speech.
Why it matters
Voice pipelines have historically stitched together separate vendors for ASR and TTS, each adding latency and per-minute costs that compound in real-time settings. A single provider tuning both ends on a common foundation — here, Alibaba's Qwen3 family — can smooth those seams.
- Accuracy: framed as competitive on Coval benchmarks
- Latency: a priority for real-time, interactive use
- Cost: positioned as a differentiator against incumbent APIs
Nari Labs is best known for Dia, its earlier open TTS work, so a Qwen3-based expansion into ASR signals broader ambitions in the voice stack. The release leaves some specifics — parameter counts, licensing terms, and language coverage — worth watching as more details surface.
Sources
- Visit
Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost
Hacker News
More from Unknown
All Unknown releases →Kandinsky 6.0 Generates Video and Audio Together
Kandinsky Lab's new foundation models produce synchronized sound and motion from a text prompt, in 3B and 29B sizes.
Liquid AI's LFM2.5-VL-DSpark targets faster VLM inference
The new vision-language model from Liquid AI is tuned for accelerated inference, extending the company's LFM2 line into multimodal territory.
ZGCM-1 arrives as a fully open 7B reasoning model
A compact foundation model targets math reasoning and agentic search with tool use, and its makers are releasing it fully open.
More in Text → Speech
All Text → Speech →Haru-4B Brings Full-Duplex Speech to Six Languages
A compact 4-billion-parameter model handles real-time, two-way voice conversation across six languages in a single system.

Zyphra releases Zonos3-Mini multilingual TTS
The compact open text-to-speech model supports multiple languages and voice cloning from short audio samples.
OpenBMB's Phonon Brings High-Quality TTS On-Device
The new text-to-speech model promises benchmark-topping voice synthesis that runs locally rather than in the cloud.
0 comments
No comments yet. Be the first to weigh in.