Audio & Speech

Today's Stories

1. 1

   NTT West's VOICENCE opens voice-AI consultation desk Sept 28, 2026

   NTT West's VOICENCE company said it will open a free "Voice Consultation Desk" on September 28, 2026. It is run with lawyer Hajime Idei and supported by Kottodori Law Office. Performers and businesses facing unauthorized AI voice cloning currently lack a clear first place to ask for help. The desk is positioned as that entry point, not a source of legal rulings.

   The desk will not give legal judgments on individual cases; it refers users to outside experts when needed. Its stated next step is publishing basic guidance on voice contracts and consent.

2. 2

   AWS Deploys Qwen3-TTS on SageMaker AI

   AWS published part one of a tutorial for deploying Qwen3-TTS on Amazon SageMaker AI using the vLLM-Omni Deep Learning Container. Because Qwen3-TTS returns audio before it has finished generating an entire response, voice agents can begin speaking sooner, cutting down on the silent gaps that make conversational voice apps feel slow.

   The sample uses a SageMaker instance pool that falls back to ml.g6.xlarge, ml.g6e.xlarge, ml.g5.xlarge, and ml.g4dn.xlarge, but quotas are required for every listed type, and the hourly rate can vary depending on the instance selected.

3. 3

   Modulate raises $25M to push audio-native AI to developers

   Modulate raised $25 million, led by Future Ventures with returning investors Hyperplane and Lakestar, bringing total backing to $60 million. The funding goes toward new software development kits and APIs, plus industry-specific models, in a push to reach developers who would otherwise rebuild audio intelligence themselves.

   Modulate says its design is up to 1,000 times more efficient than one large model, but that claim is the company's own; watch whether new developer tools convert its 10 million monthly audio hours into broader adoption.

4. 4

   Eleven v4 Released, Tops Artificial Analysis; 75% of Listeners Prefer It

   ElevenLabs released "Eleven v4" and "Eleven v4 Turbo." The v4 ranks first in Artificial Analysis's speech arena, with about 75% of listeners saying they prefer it over competitors. According to ElevenLabs, conventional speech synthesis has been limited to mechanical reading aloud, but Eleven v4 can interpret a speaker's emotions and context to generate human-like speech.

   The top ranking comes from a limited comparison by Artificial Analysis. The focus will be on Eleven v4 Turbo's response speed and the specifics of its multilingual support.

5. 5

   DetectifAI identifies AI voice on-device, eyes deals with phone makers

   San Francisco-based DetectifAI, founded by Tarini Padmanabhuni, is looking to license to handset makers an SDK that uses a small AI model running inside the smartphone OS to instantly detect AI-generated speech in calls and voice messages without sending audio off the device. Padmanabhuni argues that existing detection products run on remote servers and therefore cannot be built directly into devices by phone makers, leaving targeted individuals with few defenses; on-device detection could fill that gap.

   According to Padmanabhuni, the company already has early revenue and processes more than 100,000 calls a month for financial institutions in India, but it is not disclosing customer names, citing confidentiality agreements.

6. 6

   Google adds Live Avatar to Gemini 3.8 Live, live in Gemini Enterprise

   Google said its Gemini 3.8 Live voice model now has a Live Avatar feature, and that Live Avatar-equipped Gemini 3.8 Live is already usable in Gemini Enterprise. Live Avatar pairs live conversation with low-latency streaming video, so some customer-service and guided digital interactions can look more like real face-to-face conversations, Google says.

   Google says the avatar handles accurate lip sync, natural facial expressions and dynamic multi-language syncing across 97 languages, so the test is whether it holds that visual alignment in real support settings.