Releases·Model Release
Samsone: A Family of Open Small Audio Language Models for On-Device Inference
A new arXiv paper introduces Samsone, a family of open Small Audio Language Models (SALMs) for edge computing, with sizes of 134M, 99M, and 356M parameters. Samsone-134M sets a state-of-the-art for its size class while remaining competitive with far larger models. The team releases training code, weights, mobile-optimized checkpoints, and an open-source Android app for real-time on-device inference.
0 engagement1 sourceFri, Sep 18, 2026, 12:03 PM
AI Overview
- ArXiv paper introduces Samsone-134M, claiming state-of-the-art for its size class, plus Samsone-99M and Samsone-356M scaling variants
- Samsone models train on publicly available data; the team releases weights, training code, mobile-optimized checkpoints, and an open-source Android demo app
Entities
Related
Releases·Model Release
Mizar: A 159M-Parameter Audio-Language Model for Audio Understanding
Researchers introduce Mizar, a 159.3M-parameter audio-language model built from a CED-Small audio encoder connected to SmolLM2-135M via a frequency-merging mapper. A three-stage training recipe on ReasonAQA, AudioMCQ, and AVQA data pushes it past all prior ALMs under 200M parameters on MMAU, MMAR, and ADQA-clean. Code and checkpoints are public, and the model runs on-device with roughly one-second CPU latency.
Researchers introduce Mizar, a 159.3M-parameter audio-language model built from a CED-Small audio encoder connected to SmolLM2-135M via a frequency-merging mapper. A three-stage training recipe on ReasonAQA, AudioMCQ, and AVQA data pushes it past all prior ALMs under 200M parameters on MMAU, MMAR, and ADQA-clean. Code and checkpoints are public, and the model runs on-device with roughly one-second CPU latency.
0 engagement1 sourcearxivWed, Sep 23, 2026, 04:21 PM
Releases·Model Release
SNU researcher releases Sori-1B, a 1B audio-language model trained from scratch without text-only pretraining
Sori-1B is a 1B-parameter audio-grounded language model released on Hugging Face by a single SNU researcher. Its decoder is trained entirely from scratch on audio-paired text, with no text-only pretraining or pretrained-LM initialization, so answers ground in audio rather than text-only priors. The release includes evidence that typical audio models lean heavily on those priors.
Sori-1B is a 1B-parameter audio-grounded language model released on Hugging Face by a single SNU researcher. Its decoder is trained entirely from scratch on audio-paired text, with no text-only pretraining or pretrained-LM initialization, so answers ground in audio rather than text-only priors. The release includes evidence that typical audio models lean heavily on those priors.
6 engagement1 sourceredditMon, Aug 31, 2026, 02:48 AM
Releases·Model Release
Dev releases sanoTTS: 294k-param (337 KB) TTS stack runs on $3 ESP32; 1.5m model beats Inflect Nano and KittenTTS on SCOREQ
A developer released sanoTTS, a text-to-speech stack built while attempting to fit TTS onto a $3 chip with 512 KB of SRAM and no NPU. The release spans 294k to 2.2m parameters with 11 voices and 6 languages, with the 337 KB smallest model running fully on the microcontroller and a 1.46m-parameter model claimed to beat models 3x and 10x its size.
A developer released sanoTTS, a text-to-speech stack built while attempting to fit TTS onto a $3 chip with 512 KB of SRAM and no NPU. The release spans 294k to 2.2m parameters with 11 voices and 6 languages, with the 337 KB smallest model running fully on the microcontroller and a 1.46m-parameter model claimed to beat models 3x and 10x its size.
546 engagement1 sourceredditThu, Sep 3, 2026, 10:01 PM
Products·launched
13M parameter ASR conformer runs on ESP32-S3 microcontroller
A distilled and quantized 13.1 million parameter Conformer model (Nvidia's small Conformer) is deployed on an ESP32-S3 microcontroller costing under $10. It transcribes 8 seconds of audio using 14MB flash, 256KB SRAM, and 4MB PSRAM, though inference is slow. This enables on-device speech recognition for low-cost embedded systems.
A distilled and quantized 13.1 million parameter Conformer model (Nvidia's small Conformer) is deployed on an ESP32-S3 microcontroller costing under $10. It transcribes 8 seconds of audio using 14MB flash, 256KB SRAM, and 4MB PSRAM, though inference is slow. This enables on-device speech recognition for low-cost embedded systems.
220 engagement1 sourceredditMon, Jul 20, 2026, 04:09 PM
Developments·Paper
GigaChat Audio: Time-aware Large Audio Language Model
A new audio LLM from GigaChat introduces explicit timestamp-based question answering over recordings up to 120 minutes long. The model uses periodic time markers and synthetic supervision to achieve strong temporal grounding on both short and long benchmarks. This enables time-anchored fragment descriptions and summaries, addressing a key limitation of prior audio-conditioned LLMs.
A new audio LLM from GigaChat introduces explicit timestamp-based question answering over recordings up to 120 minutes long. The model uses periodic time markers and synthetic supervision to achieve strong temporal grounding on both short and long benchmarks. This enables time-anchored fragment descriptions and summaries, addressing a key limitation of prior audio-conditioned LLMs.
0 engagement1 sourcearxivSat, Jul 11, 2026, 04:34 PM