Audio & Speech

Today's Stories

1. 1

   Caterpillar's Cat Digital Chief Says A.I. Will Fill Labor Gap

   Cat Digital, led by Ogi Redzic, is applying A.I. across Caterpillar's equipment, including a voice-based “Cat AI Assistant” for operator coaching and development of task autonomy for machines. Facing nearly 40% of current operators set to retire by 2031, Caterpillar positions A.I. to bridge the labor shortage without displacing jobs, transitioning workers to supervisor roles.

   Whether A.I.-coordinated construction sites become common in five to ten years, as Redzic predicts, and if the productivity gains from its A.I.-built software, tripling work with only 20-30% more people, become a new standard.

2. 2

   蔦谷好位置 and Novel Core discuss 'チート'

   蔦谷好位置 and Novel Core spoke with Rolling Stone Japan about their track 'チート', focusing on classical sampling and generative AI. The discussion highlights how AI tools are influencing music production, with artists like these exploring new creative possibilities despite technical and ethical questions.

   Whether the use of generative AI in sampling becomes more accepted in mainstream music, as these established figures are already engaging with it.

3. 3

   Mitsubishi Electric develops task-agnostic sound source separation AI

   Mitsubishi Electric has developed a new technology called 'task-agnostic sound source separation' that allows a single AI model to separate multiple sound sources. This could improve audio processing in various applications, potentially leading to more efficient and flexible sound management in business settings.

   The practical impact will hinge on how this technology is integrated into real-world products and services.

4. 4

   Meta releases real-time audio model Muse Voice Transcribe

   Meta's Superintelligence Labs released Muse Voice Transcribe, a real-time audio perception model that transcribes speech, distinguishes speakers, and detects sentence boundaries during live conversation. It is available through the Meta Model API and powers voice dictation in Meta AI and Muse Code. The model undercuts rivals on price at $0.18 per hour ($3 per 1,000 audio minutes), while achieving a 3.1% word error rate on English in independent testing, compared to 3.6% for ElevenLabs Scribe v2 Realtime and 4.0% for AssemblyAI. It can handle over 20 speakers and recordings over an hour long without post-processing.

   The model's success hinges on whether its adaptive delay mechanism—which balances accuracy against speed—proves reliable in real-world use. Watch for how competitors like OpenAI, which released GPT-Realtime-Whisper in May, respond on price.

5. 5

   Google releases Lyria 3.5 music model in Gemini app

   Google has released its Lyria 3.5 music generation model in the Gemini app and via API. The model delivers more expressive vocals and richer musical arrangements than its predecessor, according to Google. Beyond the Gemini app, Lyria 3.5 is available in Google Flow Music, Google AI Studio for developers, and Google Vids. Google says the Lyria model was trained only on licensed content, unlike Suno's music model.

   The distinction in training data could appeal to businesses concerned about copyright. However, Google hasn't shared specifics about what training data it actually used, so the claim's full impact remains unclear.

6. 6

   AWS launches WhatsApp ordering assistant with AgentCore

   AWS published a solution to deploy a multimodal WhatsApp ordering assistant using Amazon Bedrock AgentCore and Amazon Nova 2. Customers can text, send voice notes, or call the same WhatsApp Business number, and an AI agent handles orders end to end. This offers restaurants a single backend and shared memory across channels, so a customer who texts today and calls tomorrow is recognized as the same person. It reduces the need to build and run separate systems for each ordering channel.

   The test is whether the cross-channel memory holds once real order volume hits, since the backend and ordering logic are decoupled from the channel. Watch whether AWS CDK deployment stays manageable as the single WhatsApp Business number scales.