Prototype to Capture Sound and Audio Attributes for Voice Analytics

Project Description

Goal:

We are a Fortune 500 company in the Healthcare sector. We want to develop software or software and a device that will capture audio and save it to a non-audio format.  We want this software/device to capture data about audio that can be used for machine/deep learning projects.  The need to protect patient privacy is the key driver of this project’s main requirement – save audio data to a non-audio file.

Our Ideal Expert:

Someone with extensive experience programming audio software and/or someone with a strong understanding of acoustics – for example, an Acoustical Engineer or someone with expertise in voice analytics.

The Requirements:

Our initial thought is to develop software that would run on a Raspberry Pi.  This is not a hard requirement, but it would be a useful starting point for a prototype.  We are open to an expert’s advice if you want to suggest off-the-shelf hardware that could be used.

The software developed will capture audio using an omni directional microphone.  The captured audio will not be written to hard disk in an audio file format.  Instead the attributes of the audio listed below, will be written to a CSV file.  Please note, the audio features listed below are only the minimum required features.  We will default to the expert’s advice for additional features to capture from a voice analytics perspective.

Features:

• Mel Frequency cepstrum coefficients
• Spectral slope
• Audio spectral flatness
• Audio spectral centroid
• Audio spectral envelope
• Any additional audio features that may help in the identification of sound

The CSV file’s rows will represent time increments (e.g. one second) and should be written as a timestamp. The file’s columns will be the audio features listed above.  An important requirement is that we as users of the software/device be able to adjust the increments for sound capture.  We also need to be able to adjust the intervals in which the data is captured.  For example, we may want to set the device to record one second of sound every two minutes.

The software/device will need the ability to connect to a Wi-Fi network and transfer data collected to a destination specified by the user. The user will also need the ability to define the interval of data transfer.

Finally, the data recorded by this device should never be able to be converted into sound (i.e. audio could not be determined from the data captured).

At the end of the project the expert should produce full specifications for the hardware device and microphone used in development.

Proposals:

In your proposal please provide the following:

1. Your suggestions on hardware and your specific approach to this problem.
2. How will you ensure that the audio data is converted to delimited file and that no audio file is created in the process?
3. What are your proposed milestones and how many hours should be budgeted for each?
4. Your relevant experience with a project like this one.