Audio & Speech

Today's Stories

1. 1

   Nvidia's free Nemotron 3 Diarization model separates 8 speakers

   Nvidia released Nemotron 3 Diarization, a free AI model with about 100 million parameters that identifies up to eight speakers in real time, detects overlapping speech, and works with both recordings and live audio. Nvidia is giving away the model weights, so any developer can add anonymous speaker labels to transcripts without paying for a proprietary service.

   Accuracy drops in noisy, reverberant settings or when more than a few people talk over each other. Watch whether the free release pushes rivals to cut prices or open their own models.

2. 2

   Kenjiro Tsuda sues TikTok over AI clone of his voice

   Voice actor Kenjiro Tsuda sued TikTok over videos he says use an AI clone of his "lustrous" baritone. The verdict is due Wednesday. It is believed to be Japan's first suit defending an individual's vocal identity against AI copies. The case is seen against a backdrop of AI imitation becoming a growing concern among performers worldwide, a threat the body says has roiled Japan's animation industry.

   Because the court has not yet ruled, the outcome hinges on Wednesday's verdict and whether a Japanese court recognizes a voice as part of an individual's identity.

3. 3

   Synthesia builds first journalist avatar, trained on one story

   Synthesia, a digital avatar startup valued at $4 billion, made an interactive avatar for a TechCrunch reporter at its New York office. It is trained only on her venture fraud story. The avatar answers press questions and runs on Synthesia's video, voice, and language models, pointing to a future where AI clones handle corporate communication and possibly journalism tasks.

   Whether avatar use spreads in corporate America hinges on trust and acceptance. The reporter predicts her Gen Z peers likely won't get used to digital twins.

4. 4

   McDonald's unveils "Archy" AI drive-thru amid sales slump

   McDonald's unveiled "Archy," an English- and Spanish-speaking AI drive-thru system it says exceeds 90% accuracy at early-test restaurants, with positive customer feedback, as it looks to revive slumping sales and a plunging stock. The launch is a bet that voice AI can reverse the chain's sales slump, even as Burger King rethinks its own AI drive-thru after backlash — a sign customer acceptance, not accuracy claims, may decide whether the format sticks.

   The test is whether McDonald's can avoid the drive-offs Burger King saw when guests didn't want to talk to a bot, per Burger King's Chakri Somisetti; watch whether the reported 90%-plus accuracy holds at full scale.

5. 5

   AWS ships Qwen3-TTS-12Hz-1.7B-Base voice cloning on SageMaker JumpStart

   AWS detailed deploying Qwen3-TTS-12Hz-1.7B-Base from Alibaba Cloud's Qwen team via Amazon SageMaker JumpStart to a managed real-time endpoint. It clones a voice from a few seconds of reference audio, in 10 languages, with cross-lingual cloning. Teams can now produce speech in a chosen speaker's voice, in any of 10 languages, from just a few seconds of reference audio — without retraining a model or collecting large training datasets.

   Whether deployments stay inside the roughly 24 GB GPU the walkthrough uses — the setup hinges on splitting GPU memory correctly between the talker and code2wav stages. Watch the KV cache token budget of 56,928 tokens for concurrency.

6. 6

   Sony, UMG sue Suno again over v6 training data

   Sony and Universal Music Group filed yet another suit against Suno, saying its v6 model is trained on user outputs from earlier models that were themselves trained on unlicensed music ripped from YouTube and other sources. Sony and UMG did not sign a licensing agreement with Suno, so the lawsuit tests whether training a new model on an older model's outputs removes the copyright problem or simply repeats it.

   The outcome hinges on whether Suno can show v6 truly started from scratch; the complaint says that unless it did, its models will continue to be built on unauthorized data.