Left is faster. Higher reads more hard parts right.
GPT-4o mini TTS is better on accuracy. Smallest Lightning v3.1 is better on non-streaming latency, streaming latency, naturalness and price.
Last updated Methodology by Dileep Chagam
Measure by measure, the same sentences, the same place.
An estimate: speech-to-text transcribers scored it, not people.
The same sentence, read by each voice.
Smallest Lightning v3.1. GPT-4o mini TTS: 688 ms. Smallest Lightning v3.1: 357 ms.
Smallest Lightning v3.1. GPT-4o mini TTS: 688 ms. Smallest Lightning v3.1: 357 ms.
Smallest Lightning v3.1. GPT-4o mini TTS: Can't stream. Smallest Lightning v3.1: 363 ms.
Neither has a figure. GPT-4o mini TTS: No stop message. Smallest Lightning v3.1: No stop message.
GPT-4o mini TTS. GPT-4o mini TTS: 82.7%. Smallest Lightning v3.1: 81.3%.
Smallest Lightning v3.1. GPT-4o mini TTS: $25.70. Smallest Lightning v3.1: $19.50.