jparedesj / voiceclass-server

Backend server for voice classification

voiceclass-server

Service for detecting sex and age category in recorded voice. The Android client can be found here

Installation

Install and run the server in a Docker container.

• First make sure to download and setup docker
• Clone this repo: $ git clone https://github.com/habbes/voiceclass-server
• Navigate to project directory: $ cd voiceclass-server
• Create docker image: $ docker build -t voiceclass .

Running the server

• Create a directory in your somewhere under your user directory where server's data files will be saved: $ cd /Users/name/some/path && mkdir voiceclass-data
• Run the container: $ docker run -d -p 5000:80 -v /Users/path/to/voiceclass-data:/data voiceclass
• The above line forwards port 5000 of the docker host to port 80 in the container. Replace 5000 with any free port on your docker host you prefer to user.

Testing the server on Linux

• From your browser/curl visit http://localhost:5000
• If it works, you will get the resonse OK

On Windows or Mac

• Find the IP of your docker host: $ docker-machine ip
• Test using the docker host IP: http://192.168.99.100:5000

API

The server exposes an API for classifying voice recordings as well training the classifier. It uses JSON for both request and response bodies.

Available Classes

• FemaleChild
• FemaleTeen
• FemaleAdult
• FemaleSenior
• MaleChild
• MaleTeen
• MaleAdult
• MaleSenior

/api/classify

Use this to get a classification of a voice recording.

Request

audio

• Type: object
• Description: Object containing the audio data and metadata used to properly decode it

audio.format

• Type: string
• Description: Audio format: currently only pcm is supported

audio.data

• Type: string
• Description: Base-64 encoding of the raw PCM data of the audio recording

audio.channelCount

• Type: number
• Description: Number of channels in the audio recording

audio.sampleRate

• Type: number
• Description: Sample rate of the recording in Hz

audio.sampleSize

• Type: number
• Description: Sample size in bytes

Response

id

• Type: string
• Description: Identifier for this classification. Used as a reference when providing feedback.

class

• Type: string
• Description: The class name for the identified sex and age category

/api/feedback

Use this to provide feedback to the system after receiving results of classification, this feedback allows the system to improve its classifier by providing additional training data.

Request

id

• Type: string
• Description: Identifier returned as part of the result of a call to the /api/classify endpoint

class

• Type: string
• Description: The correct class for the recording corresponding to the specified id.

Response

message

• Type: string
• Description: message with value "OK"

/api/train

This is used to add data to train the classifier.

Request

audio

• Type: object
• Description: Same as for the /api/classify endpoint

class

• Type: string
• Description: The class the attached audio belongs to.

Response

message

• Type: string
• Description: message with value "OK"