Skip to main content
Speechmatics is a speech recognition specialist whose long-standing focus is accent and dialect coverage rather than benchmark position on clean, standard-accent English. That focus is why it holds up on the audio that degrades generic models most. It covers 70+ individual languages, plus 7 bilingual language packs for media where two languages appear in the same file or stream.
Available in Ollang Workflows as speechmatics and speechmatics-llm — Closed Captions, Subtitle Translation, and AI Dubbing. See the Speech-to-Text catalog.

​
Models

Melia 1 is the distinctive one. It transcribes without being told the source language and follows speakers who switch language spontaneously — the cleanest answer in the catalog for content where you cannot predict, per file, what will be spoken. Its trade-off is that it does not support translation, so it is a transcription-only choice. Bilingual packs cover fixed language combinations and are the alternative when you know in advance which two languages appear.

​
Key capabilities

  • Accent and dialect robustness — consistently strong where regional accents and lesser-resourced dialects appear, which is the most common cause of unexplained quality drops in localization pipelines.
  • Speaker diarization and a custom dictionary for improving recognition of specific words and phrases.
  • Smart formatting applied by default — spoken numbers, dates, currencies, and measurements converted into conventional written forms, which is a real saving on subtitle cleanup.
  • Profanity tagging, disfluency removal, and regex-based word replacement for house formatting conventions.
  • Translation on Enhanced and Standard, with English translating to and from 34 target languages, plus Norwegian Bokmål to and from Nynorsk.
  • Flexible deployment — cloud, on-premises, or hybrid, for data-residency-constrained content.

​
Speechmatics with LLM

The speechmatics-llm variant adds an LLM post-processing pass that accepts custom guidelines and instructions. Speechmatics’ built-in formatting is rule-based; this variant adds instruction-driven judgment on top of it. Instead of correcting the same things in review every week, you state them once: speaker label conventions, fixed product terminology, number and date formatting, or client-mandated caption style. It earns its place when the alternative is making the same corrections by hand in review every week.

​
Where it fits in a workflow

Choose Speechmatics when the source has strong regional accents, when lesser-resourced dialects appear, or when speakers move between languages unpredictably and Melia 1’s automatic switching is the feature you need.

​
Trade-offs

  • Little advantage over AssemblyAI or ElevenLabs Scribe v2 on clean, standard-accent audio — its strengths show up on the material that defeats generic models.
  • Automatic language switching and translation are mutually exclusive — Melia 1 does one, Enhanced and Standard do the other.
  • 70+ languages is narrower than AssemblyAI’s 99 or Scribe’s 90+.

​
Reference