Ukrainian IT company Codebridge has created a voice application called Lispr, which allows you to convert speech to text and perform instant translation directly at the cursor location in any application. The program supports automatic detection of about 99 languages and works without mandatory registration or collection of personal data.
Ukrainian IT company Codebridge has created a voice application called Lispr, which allows you to convert speech to text and perform instant translation directly at the cursor location in any application. The program supports automatic detection of about 99 languages and works without mandatory registration or collection of personal data.
This is evidenced by the project development case study data and the official presentation of the Codebridge product.
The application operates on the Push-to-Talk mechanism. Pressing a single hotkey activates regular dictation, and a key combination launches automatic translation of recognized speech into the selected language. The recognized text immediately appears at the cursor location in any software — Slack, Figma, Terminal, text editors or development environments. At the same time, Lispr saves the data previously copied to the clipboard.
The program supports translation into 32 languages, including Ukrainian, English, German, Spanish, Japanese, and Arabic. Additional features include the ability to set up your own dictionary (industry terminology, product names, code identifiers), which is taken into account during recognition and translation.
The Whisper large-v3-turbo model is used for speech recognition, and the gpt-oss-120b model is used for instant translation. The audio stream is broadcast in Opus format (24 kbps, 20 ms packets). The optimized streaming stream provides phrase decoding with a latency of about 0.35 seconds for dictation and up to 0.7 seconds for translation. The median latency on the server is 346 ms.
The client part is implemented under macOS (a 3.67 MB native application in Swift using AppKit, AVFoundation, and AudioToolbox) and Windows (.NET 8 and WPF with a 14 MB installer signed via Microsoft Azure Trusted Signing).
The server side (Edge) is built on Cloudflare Workers (processing takes about 1 ms of CPU per dictation) using Durable Objects as a geo-relay to bypass geo-blocking of ASR providers and R2 buckets.
Lispr works without the need for registration or account creation. The audio stream is transmitted over an encrypted connection and deleted immediately after processing, without being used to train models or stored on servers.
In the first three weeks since launch, the app recorded 46,011 dictations and over 1.1 million words generated in 29 countries. The largest number of users was recorded in Ukraine (8,590 dictations), Indonesia (7,690) and Germany (1,730). A multi-agent AI system was used for development, which was involved in coding, testing and localization.
The application is distributed under a freemium model: a free version and a Pro tariff for $4.49/month are available. For comparison, the foreign analogue Wispr Flow costs $15/month and requires mandatory registration, while the system dictation of Windows and macOS does not have built-in instant translation and automatic detection of many languages. The first 1000 users can get premium access for free.
Codebridge’s future plans include releasing an application for iOS as a keyboard extension and for Android via AccessibilityService, as well as implementing a hybrid local recognition mode.
This is an IT company that provides outsourcing and software development services — from design to further support. The company was founded in 2021 by Kostyantyn Karpushyn and Myroslav Budzanivskyi.
Currently, Codebridge also uses its own products as a testbed for testing engineering practices and multi-agent AI systems before implementing them in client solutions.
