Voice agents need fast speech startup, not just natural voices
An English- and Spanish-speaking for incoming support calls and some calls to possible customers has hit an unexpected in (TTS). The is working well, while (STT) remains difficult because of noise, accents, interruptions, poor phone audio, and mumbling. Even after speech has been understood and a reply is ready, a pause before the agent speaks can make callers think the line has dropped.
Most comparisons focus on how natural a voice sounds in narration, but live calls need fast startup first, consistent speed second, and acceptable naturalness third. , Deepgram, Murf AI, Cartesia, and OpenAI TTS are being considered, with low as the main measure.
Key points
- The handles English and Spanish support calls and some calls to possible customers.
- Noise, accents, interruptions, poor phone audio, and mumbling make STT difficult.
- A long pause before the first spoken response can make callers think the call has failed.
- The priorities are fast startup, consistent speed, and then acceptable naturalness.
- , Deepgram, Murf AI, Cartesia, and OpenAI TTS are under consideration.