tensorfolk

Rime launches Prism, a fine-tunable speech-to-speech model that generates text before audio

Voice startup Rime released Prism, an enterprise speech-to-speech model that companies can fine-tune and run on-premises or in a private cloud. Because the model produces text before it speaks, businesses can check each reply against their own rules and keep audit logs. Rime also says it detects caller emotions like frustration, and Dialpad was a design partner.

Why it matters

Text-first generation offers a practical way to add guardrails and auditability to voice agents in regulated settings like contact centers.

Source: Read the full story on Rime (PR Newswire)

speech-to-speechvoice-agentsguardrailsenterprisefine-tuning

More in Generative Media

Generative Media / Voice, Speech & Voice Agents

NVIDIA opens Nemotron 3.5 Transcribe, a 30B-A4B speech model for 40 languages, plus 463k-hour Granary v2

NVIDIA published open weights for Nemotron 3.5 Transcribe, a mixture-of-experts speech model with about 4B active parameters. One model handles transcription in 40 languages, translation to and from English for 31 of them, and labeling of up to eight speakers in hour-long recordings. Alongside it comes Granary v2, a speech corpus covering 48 languages.

Why it matters: One commercially usable open model replaces separate ASR, translation and diarization pipelines, and it tops the Open ASR leaderboard among open-weight systems.

How to try it

Nemotron 3.5 Transcribe is an open-weight (OpenMDW-1.1) speech model you can run yourself for transcription, translation and speaker-labelled transcripts of recorded audio. NVIDIA's reference setup is a single H100 80GB.

  1. Set up the NeMo Speech environment as described on the model card.
  2. Download the checkpoint from Hugging Face (nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B).
  3. Load it with NeMo's ASRModel.from_pretrained and call transcribe() on a WAV file.
  4. For serving, follow the model card's vllm serve command and send audio through the OpenAI-compatible chat API.
python
import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained("nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B")
transcriptions = asr_model.transcribe(["file.wav"])
Open the docs
Big newsRead on NVIDIA (Hugging Face blog)
Generative Media / Voice, Speech & Voice Agents

Apple's iOS 27.2 brings Siri AI to five more languages and adds conversation memory

Apple shipped iOS and iPadOS 27.2, extending its LLM-based Siri to French, Japanese, Korean, Portuguese and Spanish. Siri can now recall earlier conversations and pull context from Messages photos, Calendar attendees, voicemails and Safari, while a new Excluded Apps setting limits what it can read. The upgraded Siri is still unavailable in the EU and mainland China.

Why it matters: Voice assistants beyond English reach a huge installed base at once, raising the bar for multilingual voice UX in consumer apps.

Read on MacRumors
Generative Media / Voice, Speech & Voice Agents

Block shows Dessa, a full-duplex voice model trained only on conversation audio

Block's research lab revealed Dessa, a speech-to-speech model that can listen and talk at once and was trained from scratch on recorded conversations, with no text or pretrained language model. It runs locally on a high-end Mac. It is not public yet: Block is inviting safety researchers to test it and has opened a waitlist.

Why it matters: A speech-native model with no text backbone challenges the usual speech-to-text, LLM, text-to-speech chain for voice agents.

Read on TechCrunch