Built for studios, product teams and developers. From Larchfield Agents for customer conversations to Larchfield Studio for production audio, on the speech models behind both.
Beneath the salt cliffs of Ardenmoor, the tide kept a ledger nobody had learned to read. [wryly] Not that many of them tried… [laughs] but the keeper swore the water counted back. [whispers] The gulls agreed, and said nothing.
Trusted by product teams, studios and broadcasters
Read all storiesGenerate lifelike speech, music, video and sound design.
Configure, deploy and monitor conversational agents.
Larchfield Studio
Produce natural speech, turn a script into video, score a scene, or design sound effects. Build an episode, an ad or a full audiobook without leaving the timeline.
Above the outer weather of Ardenmoor, the sky broke into fractured light, as though the coast had been glazed and hung in the air.
The sensors returned patterns no model would quite agree on.
Draft podcasts, audiobooks and voiceover in a workspace built on all of Larchfield's audio research combined.
Beneath the salt cliffs of Ardenmoor, the tide kept a ledger nobody had learned to read. [wryly] Not that many of them tried… [laughs] but the keeper swore the water counted back. [whispers] The gulls agreed, and said nothing.
Controllable, layered delivery across 74 languages and every register a script asks for.
Score full tracks instantly — any genre, any style, vocals or instrumental.
Create custom effects and ambient beds from a prompt.
Clone a replica of your own voice, design one from a prompt, or browse thousands in the library.
Create or edit stills and turn ideas into video with the Kestrel, Lumen and Torrent models.
Using synthetic narration to localise product campaigns across 21 markets
Larchfield Agents
Configure, deploy and monitor human-sounding agents in 34 languages, with high transcription accuracy and low latency across voice or chat.
Agents listen, read and reply the way people do — across phone, chat, email and messaging.
Measure resolution and satisfaction, then tune the flows that move them.
Simulate real conversations to confirm agents behave the way you specified.
Set behavioural and compliance rules that keep replies inside policy.
Handle branching conversations, apply business logic and connect systems securely.
Larchfield API
Independently rated among the leading speech models. Pick a model tuned for consistency, latency or emotional control — all of them support 31+ languages.
68ms latency for conversational use cases
Most consistent lifelike delivery
Our most expressive model yet
Our most accurate ASR model. Low cost, with speaker diarisation and character-level timestamps.
97.4% word accuracy
Studio-grade music from natural language prompts in any genre, style or structure.
Trained on licensed catalogues and cleared for commercial release
Our aim is to make communicating with software feel unremarkable. We build our own foundational models, starting with a human-like voice model and now extending well past voice.
The most accurate transcription model we have released
Mar 2026Extending research past speech into transcription, music, voice cloning, agents and more.
Learn more ›We actively monitor what gets generated with our models.
We believe misuse should carry consequences.
You should always be able to tell when audio was generated.
