Speechmatics benchmarked Melia 1 for switch-point accuracy across Arabic, Mandarin, and Tamil paired with English against Soniox, Amazon, and AssemblyAI.
Public source
Publisher name
Public post
Your users switch languages mid-conversation. 🗣️ Melia now follows those switches more accurately than any model we benchmarked across Arabic, Mandarin and Tamil paired…
Company
Speechmatics
World leading experts in speech technology. Unlocking business value in human speech using AI and Machine Learning.
- Industry
- Software Development
- Location
- Cambridge, GB
- Company size
- 51–200 employees
About Speechmatics
Speechmatics is the world’s leading expert in speech technology, combining the latest breakthroughs in AI and ML to unlock the business value in human speech. Businesses use Speechmatics worldwide to accurately understand and transcribe human-level speech into text regardless of demographic, age, gender, accent, dialect or location in real-time and on recorded media. Combining these transcripts with the latest AI-driven speech capabilities, businesses build products that utilize summaries, topics, sentiment, chapters, translation and more. Speechmatics processes over 300 years of transcription worldwide every month in 50 languages. Having pioneered machine learning in speech recognition, its neural networks consider acoustics, languages, dialects, multiple speakers, punctuation, capitalization, context and implicit meanings. Speechmatics is headquartered in Cambridge, UK with a New York office too. Speechmatics is a registered trademark.
See moreLatest activity
Latest activity from Speechmatics
5 signals
Products & Services
Speechmatics launched Melia hitting 206x real time, voice-agent latency dropped, and Linden 1 landed on LiveKit.
Presence & Recognition
Speechmatics is attending the RAI Amsterdam booth at Booth F15 from 22 to 24 September to discuss enterprise AI implementation, governance, and trust.
Presence & Recognition
Speechmatics is previewing its Melia Real-Time multilingual speech model at the IBC2026 conference in Hall 8, Stand 8.B77g.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
AudioShake
AudioShake transcribing separated tracks produces 4.1× fewer transcription errors against the strongest open-source speech separator.
Research & Knowledge
Meta
Meta achieved first place on the Artificial Analysis streaming speech-to-text benchmark with a model that has excellent latency and speaker diarization performance supporting hour-long audio input and 20+ speakers.
Research & Knowledge
AssemblyAI
AssemblyAI announced the new TRACES benchmark, which includes a Model Arena and a Harness Arena.
Research & Knowledge
Coval
Coval published benchmarks showing AssemblyAI Universal 3.5 Pro pulls ahead on accuracy at 3.0% WER.
Research & Knowledge
PolyAI