Speechmatics Speech-to-Text API (incl. Vietnamese)

Speechmatics is a voice AI company headquartered in Cambridge, UK, led by CEO Katy Wigdahl. It builds speech recognition models covering more than 55 languages, with specialised models for contact centers, healthcare, media and enterprise. Named customers include Adobe, AI Media, Content Guru and Nordhealth.

The Speech-to-Text API converts audio to text in real time and in batch using the same core models. The Vietnamese model covers standard Vietnamese and regional accents from the North, Centre and South, handles mixed Vietnamese and English speech, and adds speaker diarization and word-level timestamps. Real-time partial results arrive within a few hundred milliseconds, and voice agents can be built through integrations with LiveKit and Vapi. The models can be deployed in the cloud, on premises or on device, and translation is available as a separate product.

• Vietnamese ASR: covers Northern, Central and Southern accents and code-switching with English.
• Real-time and batch: same model for live streams and recorded files.
• Speaker diarization: separates up to 50 speakers by default, including overlapping speech.
• Word-level timestamps: supports subtitles, search and media sync.
• Voice agent integrations: works with LiveKit and Vapi for building voice bots.
• Flexible deployment: cloud, on-premises or on-device, plus a translation API.

Vietnamese banks and insurers can transcribe call center recordings on premises for quality assurance and compliance. Software companies building Vietnamese voicebots can use the real-time API as their speech recognition layer. Media and education companies can generate Vietnamese subtitles and searchable transcripts.

On the FLEURS benchmark cited on the Vietnamese page, Speechmatics Enhanced records a 7.14% word error rate versus 8.38% for Google Chirp 2, 9.32% for Amazon Transcribe and 10.08% for Microsoft Azure. One model serves both live and recorded audio. On-premises and on-device options keep sensitive call audio inside the customer's environment, backed by ISO 27001, SOC 2, GDPR and HIPAA compliance.

Voice agents, call analytics and meeting assistants all depend on accurate speech recognition, and accuracy for Vietnamese and other Southeast Asian languages has lagged English. Enterprises in regulated sectors also want speech models they can run on premises to meet data protection rules such as Vietnam's personal data protection regulations.

Headquartered in Cambridge, UK, with CEO Katy Wigdahl. Received the Queen's Award for Enterprise in 2019 and holds ISO 27001, SOC 2, GDPR and HIPAA compliance. In April 2026 it launched an on-device speech-to-text model with Adobe for Premiere, also offered as a C/C++ library for ISVs and OEMs. Founding year is not stated on the pages reviewed.