In the podcast field, the Podcastle platform recently announced the launch of its new AI text-to-speech model, Asyncflow v1.0. This innovative model provides users with over 450 different AI voices, covering multiple languages and intonations, greatly enriching voice selection. In addition, Podcastle also opens an API interface to developers, allowing them to easily integrate this text-to-speech feature into their applications, thereby enhancing the user experience.
Arto Yeritsyan, founder of Podcastle, said the company has long wanted to develop a high-quality text-to-speech model, but this goal has not been achieved due to the high training costs and data requirements in the past. However, with the rapid development of large-scale language model technology in recent years, Podcastle finally made a major breakthrough last year, and was able to build high-quality voice models without requiring a large amount of data. Yeritsyan also mentioned that Podcastle's R&D was backed by a $13.5 million Series A financing last year, which provides solid financial support for its technological innovation.
In terms of price, Podcastle's text-to-voice service is priced at about $40 per 500 minutes, which is more competitive than the $99 from rival ElevenLabs. In addition to the text-to-speech model, Podcastle's voice cloning function has also been significantly upgraded. In the past, users had to read 70 different sentences to train the pronunciation model, but now, this process has been greatly shortened to recordings that take only a few seconds. This improvement is thanks to Podcastle's Magic Dust AI technology launched last year, which significantly improves audio recording quality and makes voice cloning more efficient and accurate.
In actual testing, although the newly generated voice sounds a bit robotic, it still mimics the speaker's tone and rhythm well. Podcastle said that with the continuous advancement of technology, this feature will be gradually improved, and users can also train more natural and diverse sound effects by providing different recording samples.
Yeritsyan notes that in addition to cost advantages, Podcastle also integrates audio, video, podcast and AI-powered narrative tools into a redesigned website, a move that will set Podcastle apart from the fierce market competition. He mentioned that although most users are still mainly using Podcastle for audio content creation, the demand for video production is also gradually increasing, indicating that Podcastle is expanding its service scope to multiple fields.
Entrance: https://podcastle.ai/ai-voices
Key points:
Podcastle launches the Asyncflow v1.0 model, providing more than 450 AI voices.
The platform charges $40 per 500 minutes of text to voice, which is lower than the competitor's pricing.
The voice cloning function has been upgraded, the training time has been greatly shortened, and the user experience has been continuously optimized.