Transcription models

RiverScript offers two synchronized speech recognition models: the Standard Sync Model (OpenAI Whisper v3 + Deepgram Nova-2) and the Premium Sync Model (ElevenLabs Scribe v2 + Deepgram Nova-3). Depending on the transcription model, users gain access to different transcription capabilities. For instance, speaker diarization is available only in the Premium Sync Model.

RiverScript uses proprietary technologies for processing and filtering both incoming audio and the transcript output generated by the model. The recognition and output quality of the Standard Sync Model is sufficient for most tasks that do not require the highest level of accuracy. At the same time, our audio processing and text filtering technologies eliminate most of the issues associated with OpenAI Whisper v3 models, but they work best with high-quality audio without noticeable background noise.

However, AI speech recognition models are not perfect. The Standard Sync Model may, in rare cases, produce false matches (so-called AI hallucinations) when the audio quality is poor, especially when there is background music. In 99% of cases, RiverScript filters them out and processes them using an additional recognition model, but for tasks requiring the highest level of accuracy (such as medical audio, legal audio, and similar use cases), we recommend using the Premium Sync Model (ElevenLabs Scribe v2 + Deepgram Nova-3), available on the Premium plan. The Premium model provides enterprise-grade recognition accuracy and the best quality available on the market at the time of RiverScript’s release.

Please note that AI-powered transcription is not guaranteed to be 100% accurate and may occasionally contain errors. We recommend reviewing all transcripts that require a high level of accuracy, regardless of the model used.