Murmure

Murmure is a speech-to-text conversion application that runs entirely on the user's device. It uses NVIDIA Parakeet model to perform transcriptions without an internet connection. The system processes audio in 25 European languages while keeping all data on the local computer.

★★★★★

5.0(1 ratings)

• Local processing. The application runs all speech recognition calculations directly on the user's hardware. Audio data is never transmitted to external servers. The architecture is designed to operate without internet access after the initial installation.
• Parakeet model. It implements the Parakeet vocal recognition system developed by NVIDIA. This convolutional neural model is optimized to run on CUDA-compatible graphics cards. Transcription accuracy is maintained at professional levels for the supported languages.
• Multilingual support. Speech recognition in 25 European languages including Spanish, French, German and Italian. Automatic language detection identifies the language spoken in the input audio. Each language model occupies approximately 500 MB of storage.
• Graphical interface. Main window with recording controls and transcribed text display. The progress bar indicates the status of audio processing. Interface elements follow standard design patterns for ease of operation.
• Device configuration. Selection of input microphones and adjustment of recording levels. Audio settings allow modification of the sampling rate and capture format. Parameters are automatically saved between sessions.
• Model management. Download and installation of language models from official repositories. Integrity verification checks that model files are not corrupted. Model updates are performed manually from the settings menu.
• Result export. Saving transcriptions in TXT, JSON and SRT formats. Subtitle export includes automatic timestamps. The system clipboard receives the transcribed text with one click.
• Batch processing. Multiple transcription of audio files in WAV and FLAC format. The processing queue shows the progress of each file. Results are organized into folders according to the original structure.
• Hardware optimization. Use of GPU acceleration via CUDA for NVIDIA users. The CPU version employs AVX2 instructions for efficient processing. System memory is managed with dynamic allocation based on audio duration.
• Parameter customization. Adjustment of silence thresholds and voice sensitivity. Punctuation settings control the insertion of commas and periods. Audio filters allow for reducing ambient noise before processing.

The development of Murmure began in 2023 as a personal project. The application is written in Python using the PyQt framework for the graphical interface. The source code is available on GitHub under the MIT license. Contributors maintain the project with periodic model updates and performance improvements.

Alternatives to Murmure:

Speakey

Speakey is a real-time dictation application that processes speech directly on the user's computer, without relying on cloud services.

Price: $45   Size: 356 MB   Version: 1.3.0   OS: Windows

RocketWhisper

RocketWhisper is a desktop application for speech recognition and transcription based on the OpenAI Whisper engine.

Price: $32   Size: 110 MB   Version: 1.2.0   OS: Windows, Linux, MacOS

BB Recorder

BB Recorder is a meeting and call recording application that operates entirely on the users device.

Price: Free   Size: 22 MB   Version: 1.0.0   OS: MacOS, iOS

Vowen

Productivity software that converts speech into text and commands executed locally on macOS and Windows.

Price: Free   Size: 156 MB   Version: 0.1.12   OS: Windows, MacOS

VoiceInk

VoiceInk is a voice dictation and transcription application that uses local AI models to convert speech to text with precision, operating completely offline and respecting user privacy.

Price: Free   Size: 12.61 MB   Version: 1.70   OS: MacOS