Blynt

Blynt: Custom Speech Recognition Models That You Own

Blynt builds production speech-to-text APIs on top of open-weight speech models, fine-tuned to each customer's audience, jargon and acoustics. Customers get a streaming API tuned to their own traffic and keep the model weights, which they can run in their own cloud, on-premise or on Blynt's infrastructure.

Key Features

• Models Tuned to Your Traffic: Speech models are fine-tuned on the customer's own data so they handle their audience, vocabulary and audio conditions.

• Targeted Failure Modes: Tuning focuses on what breaks in production, such as names, phone numbers, spelled-out IDs, short yes/no answers, product catalogs and noisy calls.

• You Own the Weights: The tuned model belongs to the customer, with no lock-in to a speech API subscription.

• Deploy Anywhere: Run the model in your own VPC, on-premise, or on Blynt's cloud.

• Production Streaming: Models are served as a streaming speech API built for real-time use at scale.

• Learned Turn-Taking: End-of-turn detection is trained for the customer's own voice agent rather than a one-size-fits-all setting.

• Contextual Biasing: Domain jargon is learned at training time, and a context template lets teams steer the model at runtime.

• Measured on Your Metric: Teams tune on their data, steer at runtime and score the model on the metric that matters to them, with fixes driven by production and synthetic data.

• Fast Setup and GPU-Time Pricing: Blynt says a first model is ready in hours and production-ready in days, with cost based on GPU time at about a tenth of the running cost of closed speech APIs.

Use Cases

• Voice Agents on the Phone: Improve recognition of names, numbers, IDs and short answers on noisy calls, and tune end-of-turn detection for the agent.

• Industry Jargon: Train recognition on domain vocabulary, such as product catalogs in the food supply chain.

• Specific Audiences: Adapt speech recognition to a specific audience, such as children's voices.

• Owning the Speech Stack: Teams that want to move off a rented speech API and run their own model in their infrastructure.

Getting Started

• Website: https://blynt.ai

• Blynt is in early access: teams can request access or book a call from the website to scope a model for their voice product.

Blynt gives voice product teams a speech recognition model tuned to their own traffic, with the accuracy work done for them and the weights left in their hands.