Телеграм канал «Speech Technology»

Time to go multimodal, great dataset for audio/video things

https://github.com/FreedomIntelligence/TalkVid

Some new Turkish models (and datasets/etc)

ASR

https://huggingface.co/erayyapagci/balbal-v1
https://huggingface.co/atasoglu/seda-v0.2

TTS

https://huggingface.co/kasparovabi/drama-3-tr-speaker

Encoder might be useful, not public though

https://arxiv.org/abs/2610.31292

This paper introduces Spark-Audio 1.0, a speech foundation model trained on domestically produced computing infrastructure. The model accepts text and speech inputs and generates text outputs. It supports multilingual (99 languages) and multi-dialect (202 dialects) speech recognition, speech translation, audio perception and understanding, and both text- and speech-based dialogue, marking a shift from merely ``hearing clearly'' to genuinely ``understanding'' speech. Architecturally, Spark-Audio 1.0 couples a 0.65B-parameter dense audio encoder with a 30B-A3B Mixture-of-Experts (MoE) large language model (LLM). The audio encoder is pretrained with representation-enhanced BEST-RQ and then refined through multi-task supervised training. The audio--LLM is trained through staged pretraining, followed by post-training that combines supervised fine-tuning with on-policy distillation. In total, the model is trained on 13 million hours of audio data together with large-scale text data. We evaluate Spark-Audio 1.0 on mainstream audio and text benchmarks, as well as on in-house test sets drawn from real-world application scenarios. Despite using substantially less audio training data than Qwen3.5-Omni-Flash, Spark-Audio 1.0 achieves comparable performance on most audio and text tasks, outperforms it in multilingual speech recognition, spoken dialogue, and real-world application scenarios, and approaches the performance of larger closed-source models. Nevertheless, there remains room for improvement in instruction following and complex audio reasoning. Finally, the successful training of Spark-Audio 1.0 on domestic hardware, including Huawei Ascend 910B and Hygon BW1000 chips, demonstrates the feasibility of a domestic technical pathway for building speech foundation models.

https://huggingface.co/AuroraAI-Research/AuroraTTS-100K is another example

samples here

https://huggingface.co/AuroraAI-Research/AuroraTTS-100K/tree/main/samples

Lately there is a boom in tiny models, feels like everyone tired of growth. For example

https://github.com/yinruiqing/tiny-transducer

1M params transducer based on https://arxiv.org/abs/2101.06856

or this 1M TTS

https://huggingface.co/VantoraLabs/Vocetta2-1m

What is interesting the UTMOS reported is above 4.0 which assumes a good quality while in practice there is a clear metallic noise

https://huggingface.co/VantoraLabs/Vocetta2-1m/blob/main/samples/sample_3.wav

Feels like guys found a way to hack UTMOS getting good metrics without actual sound quality. No more UTMOS in our evaluation anymore then

Not that the TTS is great but I find that using two personas for assistant is psychologically interesting thing

https://x.com/JonPTaylor/status/2112629213211828652

Voice is somewhat robotic, not sure how it got good score

https://x.com/ArtificialAnlys/status/2112389524466929979/video/1