Soniox Speech-to-Text

Soniox offers multilingual speech recognition as an API, aimed at developers building voice agents, transcription pipelines and call intelligence products. The same models serve both real time streaming and asynchronous transcription of recorded files, across more than 60 languages, and the vendor rests its case on accuracy in difficult audio rather than on breadth of features. The capabilities that matter are the awkward cases. Language is detected automatically and the model follows switches between languages within a single sentence. Names, phone numbers, addresses, identifiers and serial numbers are transcribed character by character, which is what voice agents in support and sales actually depend on. Custom vocabulary, terminology and context can be supplied to steer recognition, and the site demonstrates that on medical drug names. Streaming adds low latency output, endpoint detection and speaker diarization as the conversation runs. Speaker recognition and translation are included rather than billed as separate services. The wider platform also covers speech generation with instant voice cloning and speech translation across 3,600 language pairs through the same API, and a consumer application is offered alongside the developer product. Pricing is usage based and published in full. Billing is calculated on tokens. For file transcription the rates are $1.50 per million input audio tokens and $3.50 per million for input and output text tokens; for streaming they are $2.00 and $4.00 respectively. The vendor states this works out to roughly $0.10 per hour for asynchronous transcription and $0.12 per hour for real time. Speech generation is billed at $4.00 per million input text tokens and $21.50 per million output audio tokens, about $0.70 per hour of generated speech. An interactive calculator on the pricing page compares these rates with other providers.