CEDAR Audio is proud to announce the immediate availability of CEDAR Cambridge 14. This is a significant step forward for the company’s flagship noise reduction and audio restoration system. Numerous important features have been unveiled in the new version, including five major processes in addition to architectural improvements.
New modules and features
- VoicEX voice extractor
- Trinity Enhance Plus
- Isolate blind source separation
- Time Stretch
- Voice Activity Detection
- Video saving with audio synchronisation
- Opus file compatibility
VoicEX voice extractor
Artificial Intelligence based upon deep neural networks (DNNs) has the potential to offer many advancements in the field of audio processing, and CEDAR’s new VoicEX voice extractor – which caused a bit of a stir when unveiled at the recent AES Conference on Audio Archiving, Preservation & Restoration – is a perfect example of this. It incorporates some unique algorithmic features that take it beyond standard DNN-based noise reduction techniques, and it’s able to separate voices (primarily but not limited to speech) from other sounds, and to perform superior noise reduction when compared with traditional signal processing techniques.
VoicEX creates two signal streams – one containing any voices identified within the input and the other containing everything else. To eliminate the noise, users need only reduce the background level to the desired degree. Alternatively, they can reduce the voice level to suppress or even eliminate the speech. It’s incredibly simple to use, and the results can be astounding.
Trinity Enhance Plus
Alongside VoicEX, the new Trinity Enhance Plus filter bank is derived from the latest CEDAR Trinity surveillance system and is designed specifically for audio forensic use. It provides six related tools that allow users to suppress background noise, increase the intelligibility of speech, and enhance the background if that contains wanted detail. With just six controls, it offers huge flexibility but remains simple enough for novices to achieve excellent results across a wide range of material.
The human auditory system has a remarkable ability to focus its attention on one element in a noisy environment, which is why it’s possible for you to have a conversation in a noisy space such as a club or a café. Unfortunately, some of the information that enables the brain to perform this task is lost when the audio is recorded. This means that audio recording and surveillance in noisy environments is often unsuccessful. To overcome this, CEDAR’s unique Isolate blind source separation module identifies and separates individual audio sources within a live stream or recording obtained using a suitable microphone array. It can identify all of the independent sound sources within the soundfield (not just the loudest ones), and it will work with any suitable multi-channel signal, whether the geometry of the array is known or not.
This is very different from beamforming, which is a technology for steering a virtual microphone in a known direction. In contrast, Isolate decomposes the unknown soundfield into multiple distinct sources of interest without a-priori knowledge of the locations of the speakers or other sounds. Furthermore, the microphone array can be far smaller than that needed for a beamformer. For both of these reasons, Isolate is more flexible and can achieve superior results quicker and more easily than other methods.
Although it has uses elsewhere in CEDAR Cambridge, the new Mixer module was designed specifically to allow users to monitor selected signals obtained from Isolate, allowing them to mix wanted sources to identify and reconstruct conversations from its multi-channel output.
Sometimes you may want to alter the duration of some or all of an audio clip without affecting its pitch. This can be particularly useful when trying to increase the intelligibility of speech, so the algorithm in Time Stretch has been optimised for speech signals. Users can adjust the speed of signals within the range 20-200 percent without altering their pitch, and Time Stretch offers four modes to preserve the quality of different types of signal. Users can even apply time stretching to AV files. Not only will the audio be stretched, but the video will also be stretched to maintain synchronisation.
Voice Activity Detection
The Voice Activity Detector is another area in which AI makes its mark in the new system. Listeners working with large volumes of material will often want to focus on sections of the audio that contain speech and, up until now, this was only possible by listening to the recording. In CEDAR Cambridge 14, a DNN indicates where it is most likely that speech occurs in the timeline. This will be a major benefit to users working in the field of audio forensics.
Video saving with audio synchronisation
Another major new feature will benefit everyone who works with audio extracted from video files. Previously, the processed audio had to be saved to an audio file and recombined with the video in a video workstation, but CEDAR Cambridge is now able to save video files with the cleaned audio perfectly synchronised with the video – even if the audio has been time stretched or processed using Respeed to remove problems such as wow and other speed variations. It’s not just that the correct duration is maintained – the video is correctly rendered at every point, tracking the speed variations to retain lip-sync throughout.
Opus file compatibility
CEDAR Cambridge 14 is now compatible with the Opus audio file format. Originally developed for efficient internet communications, this has been widely adopted by manufacturers working in the fields of audio forensics and surveillance and is widely held to offer superior audio quality to other compressed audio formats at any given bit-rate.
Everything contained within this announcement is immediately available. There’s no waiting for months for CEDAR Cambridge 14 to become reality.
Free updates for existing users
As always with CEDAR Cambridge, existing users can upgrade free of charge and obtain free updates for any previously purchased processes. Users should contact firstname.lastname@example.org for download information.