Speechmatics: Uses, Pricing and Evaluation Guide
What is Speechmatics?
Speechmatics provides batch and real-time speech APIs, including mixed-language recognition and conversational turn detection. Evaluate language changes and domain terms on your recordings.
Pricing and access
Melia 1 is listed at $0.24 per hour for recorded audio and $0.40 per hour for real-time audio. The page offers $100 starting credit. Other models and optional discounts have different conditions; list rates are shown before training and volume discounts.
Who should shortlist it?
Teams processing multilingual recordings can compare batch and streaming paths. Developers still implement uploads, job management and downstream review. Select the model for the actual language behavior rather than the largest advertised language count. A batch archive and live captions need different latency and recovery expectations.
Build a mixed-language trial
Include mid-sentence language changes, a product name resembling an ordinary word and overlapping speakers. Have a qualified reviewer annotate the reference. Score switching, numbers and speaker boundaries independently. A sensible translated paragraph does not prove the original words were recognized correctly.
What should survive processing?
Retain timestamps, speaker uncertainty and qualifications. Include silence, clipped audio and interrupted streams. Check whether retries duplicate text or shift timestamps. Review the transcript before a summary, because fluent summarization can conceal recognition errors and make omissions harder to locate.
Cost and data decisions
Multiply model rates by processed hours and add retries and downstream analysis. Training-related discounts are data-policy decisions, not just cheaper prices. Verify account region and retention requirements. Specialized model labels alone do not establish suitability for a high-stakes workflow; use representative audio and a qualified reviewer.
Alternatives and selection
Compare AssemblyAI and Deepgram with identical audio and output requirements. Calculate cost per corrected usable transcript, not merely per uploaded hour.
Sources checked October 30, 2026: Official product and pricing information.