Release Notes

Current deep learning model: v2.40.0(Jun 01, 2026)

September 19, 2026

Cochl.Sense Edge SDK v1.6.1 released.

  • add native license backend (now the only backend): online activation with automatic heartbeat renewal; C++/Python use a local daemon, Android runs in-process (no daemon)
  • add default_hopsize option to config.json (inference hop interval, in seconds)
  • add runtime sensitivity by name, matching config.json (SetSensitivity, per-tag SetTagSensitivity)
  • reduce shipped libsense-core size

Breaking changes

Since v1.6.1, the Cochl.Sense Edge SDK uses a native license backend and a processor-based API, and several config.json keys have changed.

  • License: the native license module replaces Thales. On C++ and Python, you should install the license daemon once (sudo ./sense/sl/scripts/install_daemon.sh); Android needs no daemon.
  • API: C++ and Python now build a processor and receive results in a callback (sense::AudioProcessorBuilder, sense.create_file_processor, sense.create_stream_processor). Init now takes the contents of config.json instead of a path. On Android, predict returns every completed window at once as { "frames": [ ... ] }. Each platform tab on the Sample page lists the removed calls and their replacements.
  • config.json: app_path has been renamed to app_data_path, and default_hopsize has been added. log_level and every sensitivity value are now strings ("error", "warn", "info", "debug"; "VERY_LOW" … "VERY_HIGH"). A v1.6.0 config still loads, but values it no longer understands are dropped silently: a numeric log_level becomes "info", a numeric default_sensitivity becomes "NORMAL", and numeric per-tag entries and the removed model_delegate are discarded. So you should convert them before upgrading.
  • Defaults: num_threads -1 → 0 (both mean all cores), metrics.retention_period 7 → 0 (local metrics saving is now off by default), metrics.free_disk_space 500 → 100, metrics.push_period 60 → 30.
  • Result format: values are now JSON numbers instead of strings, and prediction_time has been renamed to prediction_time_ms.
  • Sample rate: audio below the model rate (22,050 Hz) is now rejected: file mode fails with a resampling error, and stream mode rejects the pushed chunk. You should capture at or above model_sample_rate.

See The Sample Page for more information.

June 1, 2026

New deep learning model v2.40.0 has been released for the Edge SDK.

  • approximately 17% AP improvement over the previous release.
  • the biggest accuracy gains are on small / hard-to-detect target sounds and noisy environments.
  • no SDK API changes — existing apps pick up the new model on next initialization.

August 19, 2025

Cochl.Sense Edge SDK v1.6.0 released.

  • add config.json for SDK configuration
  • rename feature: “Result Abbreviation” → “Result Summary”
  • add new features: “Audio Activity Detection”, “Automatic Gain Control”
  • add supported platforms: Python 3.9, 3.10, 3.11, 3.12; Android NDK 26 (API level 34+)
  • add supported hardware: GPU on JetPack 4.6.5 and 5.1.3 (Compute Capability 6.x, 7.x, 8.x)

SDK for GPU is not available on Resources page. If you need, please contact support@cochl.ai.

Breaking changes

Since v1.6.0, the Cochl.Sense uses json file for configuration.

See Advanced Configuration page for more information.

November 13, 2024

Cochl.Sense Edge SDK v1.5.0 released.

  • remove features: “Default Hop Size”, “Loudness Filtering”
  • fix issue with license renewal
  • update audio samples
  • support multi-second model

Breaking changes

Since v1.5.0, the Cochl.Sense uses a new model with a 2-second window size and a 1-second hop size.

The format of the model’s metadata has been changed. So you should delete the models you’ve been using before. You can download a new model file after removing the .sense-sdk directory (the default path is ~/.sense-sdk).

A new public method for AudioSourceStream has been added (get_hop_size and get_window_size).

December 28, 2023

Cochl.Sense Edge SDK v1.4.0 released.

  • fix issue with gradually increasing memory usage
  • add new features: “Default Hop Size”, “Sensitivity Control”, “Result Summary”, “Loudness Filtering”
  • apply new licensing method

Breaking changes

Following the new licensing, there have been changes to the authentication process.

See The Sample Page for more information.

June 21, 2023

Cochl.Sense Edge SDK v1.3.0 released.

  • add “log_level” field for log-level configuration
  • optimize metrics processing
  • fix error when running on armv7l platform
  • fix issue where deleting an SDK channel on the Dashboard was not reflected

September 28, 2022

Cochl.Sense Edge SDK v1.2.0 released.

  • require audio source file as a file path (.wav or .mp3) instead of a buffer
  • downsample audio if sample rate > 22,050 Hz
  • add “device_name” field to identify device on the Dashboard
  • remove service concept

Breaking changes

Following the new no-service concept, the model format has been modified.

July 07, 2022

Cochl.Sense Edge SDKK v1.1.0 released.

  • support multi-tag
  • apply new logging format
  • fix memory leak related to metrics
  • reduce CPU and network usage
  • add metrics-related parameters
  • improve pushing local metrics to server on termination

Breaking changes

Following the new multi-tag feature, the metric format has been modified.

See Upload SDK Analytics for more information.

March 10, 2022

First release of Cochl.Sense Edge SDK for Python (v1.0.0).

March 02, 2022

First release of Cochl.Sense Edge SDK for C++ (v1.0.0).