Topic
Technology that converts spoken audio into text, used in transcription, voice assistants and conversational AI, spanning many languages and accents.
IBM says its Granite Switch 4.1 8B and 30B models run chat, a decision adapter, a guardian adapter and speech recognition in about one base model's VRAM. The memory claim is IBM's own, so any GPU saving depends on what a team serves today.
Reality
NatWest and Nvidia cut misheard words 20% in the Spending Insights voice tool, now open to 75% of Royal Bank of Scotland customers. The announcement did not disclose a starting error rate, so the gain cannot be sized in absolute terms.
Reality
Interfaze released interfaze-1-lite, an Apache 2.0 open-weight model for OCR and speech that returns confidence scores and bounding boxes with its answers. Teams can send doubtful rows to a reviewer once they have tested how the scores behave on their own documents.
Reality
Microsoft's MAI-Transcribe-2 covers 60 languages with speaker labels and word timestamps, though streaming is not among its documented features. Live voice products get a fast model for replies and still need another way to hear the caller.
Perspective Coverage
8 publishersReality
David AI's 21-language benchmark found speech systems' error rates spread an average of 25.4 points in five Indic languages, against 3.7 in the tightest five. In those languages, choosing a system is a 25-point decision that an English score can make look like a close call.
Reality
Modulate raised $25 million to put models that analyze raw call audio, not transcripts, in front of more developers. Its pitch to teams running voice agents is a separate layer that spots cloned voices and grades how agents handle callers.
Perspective Coverage
3 publishersReality
Gemini 3.5 Transcribe ships as a sub-second streaming endpoint and a batch one, with filler-word removal and formatting inside the model. The catalog and the pricing page have not caught up.
Perspective Coverage
6 publishersReality
NVIDIA reports a 14.72% diarization error rate for the 100M-parameter Nemotron 3 Diarization, first on Voice Arena's initial leaderboard. Its Hugging Face preview page confines the weights to internal testing and evaluation.
Reality
The author of a Crimean Tatar recogniser scored it for the first time at 34.6% word error, then found four of his audiobooks sitting in the data twice under filenames a name check had cleared.
Reality
An RTX 3090 holding transcription and embedding models in VRAM averaged 25 W over the month at a cost of 2 euros, while the same post puts hardware amortisation at about 25 euros a month, twelve times the power bill.
Reality
A developer's write-up of a medical transcription pipeline cites a 2025 study finding that under noise Whisper puts confidence above 0.7 on tokens that are wrong, and that the overconfidence grows as the signal-to-noise ratio falls.
Reality
Apple's Kids Category rule and COPPA's definition of a child's voice both point a preschool voice product at local processing, and Whisper's error on children's speech falls by nearly a factor of three from tiny to large-v3.
Reality
Granite Speech 5.0 Turbo CTC drops the language model and three quarters of its output steps. The speed looks real; the accuracy figures are IBM's own.
Reality