IIITB Janastu Mphasis on Intentional Audio Networks

Sharat has joined IndicNLP project as of yesterday. We can go ahead and plan the next steps. Shall we start off with going through the datasets and formulating one or more smaller problem statements?

June 3rd Minutes of Meeting (IIITB team meets Janastu team)
https://open.janastu.org/daily/minutes/2021/june2021#03-june-2021 (see links referred)

** 3 repositories to share: **
@mani to help oversee sharing with Sharath online repos from harddrives

Set 1

Anthillhacks 2015 and Nam Halli Radio 2019-2020

​​​​[ ] da.pantoto.org and dash.swtr.us and papad
​​​​[ ] scrape namdu1radio.com

http://da.pantoto.org/
http://dash.swtr.us/#home
https://www.namdu1radio.com/
http://papad.pantoto.org/

Nam Halli Radio - Covid Campain Audio
https://drive.google.com/drive/folders/1MC4A00umYHqRA9nF13W-eM3Ro-H0DAWT?usp=sharing

Nam Halli Radio - Garikur Radio Neelgiris
https://drive.google.com/drive/folders/1cjD9hD1RvJYNjZOdWr_SpoY9goO1ynrS?usp=sharing

Nam Halli Radio - Indoansia radio files
https://drive.google.com/drive/folders/1dacPFP6VfN-9U_Q-GVdr-CQj1OEieyTV?usp=sharing

Nam Halli Radio covid19 campaign
https://files.janastu.org/s/KbGd5RJGswsNna5

AnthillHacks2019 audio https://drive.google.com/drive/folders/1LtflOdWet3RrAEonzh9b2bK5XnCeMZog?usp=sharing

Set 2

Mirzapur girls activity data (gig enabling)

​​​​[ ] provide links and fragment tagging sheets @shafali @madhu ?
​​​​[ ] Nextcloud data related to this

humani pehchan recording
https://docs.google.com/document/d/1K1HBTag8QP7Gc8cGeGdw5_1OG0ltWqOSUxCRLAbiXlk/edit

mirzapur- Papad audio annotations data
https://docs.google.com/spreadsheets/d/1aHN2NAQLKTQnlrDUnn7Ri6VkOR-RnDtNbALNfEqeB1k/edit?usp=sharing

Village Diaries, Songs, Stories, Videos
https://docs.google.com/document/d/1AgdW-T3d08GsChkdJb1dBqFT-orGCbpsvIV67uyHoVg/edit?usp=sharing

Mirzapur-Activity
https://files.janastu.org/s/6sJQT2Nrd95pcAy

Mirzapur Girls: humare game plan https://docs.google.com/document/d/1u5U7fQft4OhIhEgqy1j1LSCNGXY2jJj4ISQAiCALc2g/edit
Content Collected from GG
https://docs.google.com/document/d/1AgdW-T3d08GsChkdJb1dBqFT-orGCbpsvIV67uyHoVg/edit
Humanre Sapne https://docs.google.com/document/d/1wq1TOcJ4khVgJ26GbdycQrgz6AnHSN0SZ95zUCUvgmM/edit
Drive link https://drive.google.com/drive/u/0/folders/1BEaaE5rfIWPac9bQEGKnSXLBnc2xOBSH

NCBS stories from Archives

​​​​[ ] pointers to oral histories (see sandbox)

13ways-sandbox-oral-history: http://stories.archives.ncbs.res.in/exhibit/13ways/#/theme/sandbox/oral-history

Sprint plan

Sprint-1: 09th to 20th Aug, 2020

Minutes of meeting

02nd July 2021

(Sharath had a discussion with Prof Ram students who would help us in ASR.)

Following are the points from discussion:

1. Sharath will share audio corpus with only conversations and announcements. Audios with music will not be included.
2. Also corpus will only contain audios with Kannada words.
3. Dhanya and Tirthankar will generate Kannada and English transliteration texts from shared audio corpus and Sharath will validate the text.

To improve the accuracy of conversion following is planned:

5. Dhanya will create segments of audio from corpus audio files and share with Sharath.
6. Sharath would listen to audio segments and create transcriptions segment wise.
7. Dhanya and Tirthankar will use the transcripts to train the ASR model and improve accuracy.

Sharath will share the corpus with points 1&2 by 6th July.

21st June 2021

(Indic NLP Dataset and Concepts Discussion with Sharath, Dinesh & Mani)

Janastu works can be broad classified as:

1. Creating informational content and renarrations for classes of people who do not have minimmal academic background.

This is achieved by Alipi and SWeets. Following is the explanation of how it works: For given content the renarrations are compiled by annotations or SWeets for sections of the content. SWeets are semantic annontations for sections of text written in languages like English. SWeets have target languages. Alipi is a platform to aggregate SWeets and compile renarrations. Alipi aggregates all SWeets of target language for a content and creates a renarration in the target language. The SWeets can be text, audio or video contents in the target language. SWeets scheme was replaced by W3C Annotations Scheme.

2. Semantically linking the content created by classes of people who do not have minimal academic background.

The scope of the project "IIITB Janastu Mphasis on Intentional Audio Networks" is to link the audio contents. We discussed that, we would process the audio content, interlink audio contents based on semantics, n-grams, etc. The interlinking of audios is open problem and Sharath would be working on it. For initial analysis and study we have picked "Nam Halli Radio - COVID Campaign Audio" corpus (https://drive.google.com/drive/folders/1MC4A00umYHqRA9nF13W-eM3Ro-H0DAWT?usp=sharing)

28th July 2021

Indic NLP Technical Discussion (Prof. Sridhar & Sharath)

1. In the corpus we have chosen, 60℅ audios have ambient noise, background audio, etc. Try audacity or any other tool and create audios with clear speech content. Speech of men normally has 3KHz frequency and higher for women and children.
2. Experiment audios content with Google ASR functions: Listen, Recognize, Transalate, etc. to speed up annotation.
3. Dhanya has shared segments of audio and current ASR output. Listen to audio and refer to generated text and create clear text to be used in training.
4. Will do the annotation for samples and share by this weekend and ask for required samples to annotate for good training and complete annotation by 9th August.