Archive — US — techcrunch
OpenAI upgrades its transcription and voice-generating AI models
OpenAI has upgraded its transcription and voice-generating AI models, introducing gpt-4o-mini-tts and gpt-4o-transcribe. These models enhance the user experience with customizable voice emotions and improved accuracy in speech recognition. Unlike previous models, they won’t be available open source, reflecting a shift in OpenAI's approach amid increasing demand for tailored AI solutions.
Key points
- OpenAI launches new transcription (gpt-4o-transcribe, gpt-4o-mini-transcribe) and voice models (gpt-4o-mini-tts).
- The new voice model offers nuanced speech that can reflect emotional tone based on context.
- Transcription models are claimed to handle varied and accented speech better than the older Whisper model, which often hallucinated inaccuracies.
- The new models have a reported word error rate close to 30% for certain Indian languages, indicating room for improvement.
- OpenAI opts not to release the new models as open source due to their size and complexity, opting for a more cautious approach to deployment.
Why it matters
- As AI technology evolves, companies are increasingly focusing on creating tailored, context-aware applications that improve user interaction—reflecting trends in user experience and AI ethics seen in 2025.
- The move to not release models as open source aligns with broader industry shifts toward controlled access to powerful AI technologies, which has been a point of debate among experts.
OpenAI's advancements signal a significant shift toward developing more responsive and tailored AI solutions, reflecting a broader trend to prioritize controlled deployment and user-centric design in future AI innovations.