Audio & Speech

Today's Stories

1. 1

   Cohere releases open-source Arabic speech-to-text model

   Cohere released Cohere Transcribe Arabic, a 2-billion-parameter open-source model for Arabic speech recognition. The model is available on Hugging Face and through the Cohere API under the Apache 2.0 license. According to Cohere, it is the most accurate open-source Arabic speech-to-text system available and outperforms Whisper Large V3 and the standard Cohere Transcribe model in benchmarks. It addresses Arabic's specific challenges—dialect variety, bilingual Arabic-English conversations, code-switching, and specialized vocabulary—which are difficult for general speech recognition systems to handle accurately.

   Human ratings on a 1–5 scale show Cohere Transcribe Arabic outperforms both Whisper Large V3 and the standard Cohere Transcribe model in overall quality, dialect faithfulness, and code-switching. The model is available now on Hugging Face and via the Cohere API.

2. 2

   Anthropic clashes with Trump's White House, rejects Washington playbook

   The Trump administration has twice taken actions against Anthropic—labeling it a "supply chain risk" in April after the company refused Pentagon contract language, and imposing export controls on its Mythos and Fable AI models two weeks ago following discovery of a jailbreak. OpenAI, by contrast, announced it was withholding release of GPT-5.6 at the U.S. government's request on the same day those controls were relaxed. Anthropic, valued at $965 billion（約150兆円） and preparing for an IPO expected in the coming months, has refused the flattery, donations, and appointment of Trump allies that other tech giants (Meta, Amazon, Apple) and OpenAI have used to stay in the administration's favor. Trump administration officials have publicly attacked CEO Dario Amodei as a "liar" with a "God-complex" and an "ideological lunatic," and accused the company of "regulatory capture." Continued hostility could make it harder to sell public market investors on the stock listing and significantly hamper the company's ability to develop advanced AI models.

   Anthropic CEO Amodei reportedly called Trump "a feudal warlord" in a now-deleted Facebook post, and his sister and cofounder Daniela Amodei donated to Kamala Harris's campaign. Unlike OpenAI's policy chief Chris Lehane and cofounder Greg Brockman (the largest donor to Trump Super PAC MAGA Inc.), Anthropic has made no similar hires of Trump-aligned figures to its leadership.

3. 3

   Netflix uses AI-generated Gene Wilder voice for Wonka reality show

   Netflix is premiering Wonka's The Golden Ticket on September 23rd, a reality competition based on the fictional Wonka universe. The show's voiceover uses an AI-generated version of Gene Wilder's voice, created in partnership with AI audio company ElevenLabs and with consent from Wilder's family. This extends Netflix's pattern of using AI-generated celebrity voices for content—the company has previously recreated voices of Michael Caine and Stan Lee. For viewers, it means encountering synthetic versions of iconic figures in new productions, blurring the line between archival and synthetic media in mainstream entertainment.

   The two-part finale airs on September 30th. The show will feature 12 golden ticket winners and their chosen partners competing in a high-stakes social experiment, with one champion crowned by the end.

4. 4

   Reddit User Struggles to Replicate Pocket TTS Implementation

   A developer attempting to implement Pocket TTS (a text-to-speech model from Kyutai Labs) on smaller datasets reported that despite low training loss metrics, the model fails to generate meaningful speech even on text from its training set. The model was trained on LJSpeech and LibriSpeech clean subset, achieving flow matching loss near 0.20 MSE and very low EOS loss, but inference at epoch 2800 produced poor results. The case highlights a common gap between training metrics and real-world performance in machine learning implementation. When researchers do not release training code, developers attempting to reproduce work from scratch face a reproducibility challenge — even with low reported loss, the model may not work as expected, which can signal issues in architecture choices, data preprocessing, or hyperparameter tuning that are not obvious from the paper alone.

   The developer tried scheduled sampling and Gaussian noise injection to address exposure bias and hallucination, but neither resolved the core problem. The underlying cause — whether architectural, data-related, or optimization-based — remains unidentified in the post, suggesting that community input or access to the original training pipeline would be needed to debug further.

5. 5

   Hacker News user seeks AI voice narration better than ElevenLabs

   A user posted on Hacker News asking whether any AI voice narration tool surpasses ElevenLabs, citing dissatisfaction with ElevenLabs' flat, inconsistent character voices for animation work. The question reflects ongoing frustration with current AI voice generation quality for creative projects—the user previously relied on Seedance 2.0's 15-second voice reference feature but found it limited, and spent $75 on a human voice actor whose work was unusable.

   The post received no comments at the time of publication, meaning no community response or recommendations were available yet to address the user's request.

6. 6

   NagaTranslate builds translation and speech tools for India's low-resource Naga languages

   A developer has built NagaTranslate, a translation and speech pipeline for Nagaland's low-resource languages—currently supporting Nagamese, Ao, and Sema. The project uses commercial LLM APIs with optimized prompts for text translation, after starting with a fine-tuned NLLB (No Language Left Behind) model. Nagamese and other Naga languages were historically oral with very little standard parallel data, making this an open challenge in low-resource NLP. The project demonstrates a practical approach to building AI tools for languages that typically receive minimal tech investment.

   The developer is sharing the architecture and seeking feedback on how to improve the pipeline under strict resource constraints, suggesting the work is still in active development.