whisper

Reshaping Education using AI powered Speech Processing for Underprivileged Youth

Souvik Datta

• 6 min read
Reshaping Education using AI powered Speech Processing for Underprivileged Youth

Unlocking Infinite Possibilities in Language Learning: Nudge & Monster API's Symphony of Innovation

India is a diverse nation, home to numerous languages, dialects, and cultures. Within this diversity, English holds a vital position, acting as a tool for socioeconomic advancement and global communication.

The/Nudge is an institute with a mission to alleviate poverty within our lifetime. One of the three verticals being Center for Skill Development and Entrepreneurship (CDSE) runs a program called - Future Perfect.

Breaking Down The Walls: The Hurdles in the Path

Improving English skills among underprivileged youth has been a challenging journey. One significant obstacle was creating an environment for students who lacked practice partners to receive constructive feedback on their speaking skills. The goal was to develop a virtual learning platform where spoken words could be accurately analyzed, and feedback could be provided smoothly.

The Transformative Technology: A Whisper of Change in the Wind

In their effort to achieve a substantial impact, The/Nudge came across Monster API's Whisper ASR System. This advanced Automatic Speech Recognition (ASR) system not only made their ambitious educational model a reality but also fulfilled their crucial need for affordable speech-to-text conversion, which is vital for supporting The/Nudge's non-profit mission.

MonsterAPI's ASR system, with its diarization capabilities, could differentiate speakers even in group discussions, transcribing dialogues with precision. This allows for a careful analysis and critique of each student's speaking patterns, language use, and interactive skills.

What truly distinguishes Monster API is its exceptional cost-effectiveness and top-tier performance. For just $1, it can transcribe approximately 330 minutes of audio. Compared to alternatives like OpenAI and AWS, it offers an impressive 80% cost reduction, ensuring financial sustainability for The/Nudge.

To highlight the significant cost advantage of this technology, consider the following:

  • Transcribing 10,000 hours of audio with Monster API costs just $1300.
  • In stark contrast, transcribing 10,000 hours of audio with AWS Transcribe carries an exorbitant price tag of $11,250.

This results in a massive savings of nearly $10,000 when choosing Monster API, all while maintaining the highest levels of accuracy attainable with Whisper Large-v2.

Currently, The Future Perfect program serves 100 students per month and aims to increase this number to 200 in the near future, with expectations of doubling this figure within the current quarter. With the adoption of Monster API, The/Nudge can now serve 11 times more students than previously feasible, resulting in significant savings that can be redirected toward expanding their educational initiatives and positively impacting even more lives.

The infrastructure behind Monster API, leveraging a vast decentralized network of GPUs, contributed to its exceptional cost-effectiveness. This GPU infrastructure coupled with an optimized model ensured a 12x improvement in cost efficiency while consistently delivering top-tier performance.

In their collaboration with Monster API, The/Nudge not only found a budget-friendly solution but also a powerful tool that amplified their mission. This combination of cost-effectiveness and high performance showcases how technology can be a driving force for positive societal change.

Behind the Scenes: Where Technology Meets Education

The collaboration between The/Nudge's program and Monster API's technology resulted in "Ukti," an innovative learning module where students engage in structured speaking rooms.

Here's a simple breakdown of the process:

  1. Engage and Record: Students have conversations on specific topics in breakout rooms. Future Perfect (FP) bots record the discussions while ensuring privacy by keeping participants anonymous.

2. Listen and Transcribe: Monster API transcribes these conversations while distinguishing between speakers, ensuring accurate, individualized feedback.

Transcript example:
Speaker 0: I we're choosing the first topic .
Speaker 1: Mhm . Our topic is first how they um one second ,
Speaker 0: I'm sorry . Uh Central market order system .
Speaker 1: So me up .
Speaker 0: Yes . Yes . Our topic is not taking space .
Speaker 1: I , I will take the screenshot . Yes . Our topic is was uh
Speaker 0: ok . Ok . Ok . He is talking in uh ,
Speaker 1: yeah ,
Speaker 0: you got no apology . Yes . Yes . Yeah , I
Speaker 1: would like to apologize , apologize . No . Uh I won um because I um , I am only uh talking to uh a breakout room . Uh You
Speaker 0: are only attacking in the breakout room . Yeah , you do not give chance to other member .
Speaker 1: Uh Yes . Hi Jay . Uh I will write down . Yes . Um because we continue talking . Hm .
Speaker 0: And continuously talking
Speaker 1: me , we contain famously talking ,
Speaker 0: talking in my record room . So kindly . Sorry .
Speaker 1: Oh , or uh so sorry ,
Speaker 0: friendly , sorry about that . And I will not do next time . This . Hm .
Speaker 1: Yeah , I uh please uh I don't uh repeat it the next time . Uh Sorry for guys . Uh sorry to say this guys . Uh please uh uh forgive me .
Speaker 0: Yes , the reason .
Speaker 1: Mm because I like the topic . So I feel so excited to talking
Speaker 0: continuously . Yes . Yes . Actually , you , you're doing the , that , that much and he was , that's fine . Sorry . Can you actually doing that , that song ? You are continually talking in bracket room giving to others the chance to talk .
Speaker 1: You tell to me
Speaker 0: you are actually
Speaker 1: work from home
Speaker 0: and , you know , work from home . Uh ,
Speaker 1: how do I have , uh , ok .
Speaker 0: You know , today , say it again .
Speaker 1: Uh , New President ,
Speaker 0: an Animal .

3. Analyze and Educate: Transcripts are then thoroughly analyzed using LLMs (OpenAI), identifying errors, suggesting corrections, and providing explanations.

Speaker 0:
Identified Incorrect Statements:
1. "I we're choosing the first topic." (100)
2. "He is talking in uh, you got no apology." (100)
3. "Can you actually doing that song?" (100)

Provided Corrections:
1. "I am choosing the first topic."
2. "He is talking without giving an apology."
3. "Can you actually do that soon?"

Provided Explanations:
1. The verb "we're" is incorrect. It should be "I am" to indicate the present tense.
2. The phrase "uh, you got no apology" is grammatically incorrect and unclear. It should be "He is talking without giving an apology."
3. The phrase "Can you actually doing that song?" is grammatically incorrect. It should be "Can you actually do that soon?" to form a proper question.

Speaker 1:
Identified Incorrect Statements:
1. "Our topic is first how they um one second, So me up." (100)
2. "And continuously talking, me, we contain famously talking, talking in my record room." (100)
3. "Sorry for guys." (25)

Provided Corrections:
1. "Our topic is how they attack in the breakout room."
2. "And continuously talking, we keep talking in the breakout room."
3. "Sorry, guys."

Provided Explanations:
1. The phrase "how they um one second, So me up" is grammatically incorrect and unclear. It should be "how they attack in the breakout room" to express a clear topic.
2. The phrase "continuously talking, me, we contain famously talking, talking in my record room" is grammatically incorrect and unclear. It should be "And continuously talking, we keep talking in the breakout room" to express a clear statement.
3. The phrase "Sorry for guys" is grammatically incorrect. It should be "Sorry, guys" to express a proper apology.

4. Feedback with Care: Finally, personalized feedback is delivered directly to students via WhatsApp, supporting continuous learning outside virtual classrooms.

Immediate Learning Impact: Ukti's Feedback Loop

The true value emerges when technology fades into the background, allowing learning to take center stage. Ukti offers students a safe space to practice, make mistakes, and, most importantly, learn from those mistakes with prompt, personalized feedback. This continuous cycle of practice, correction, and improvement ensures long-lasting learning and progress.

Building Futures, Amplifying Voices: The Power of AI

The partnership between The/Nudge and MonsterAPI goes beyond just using technology; it represents what happens when a ‘purpose meets innovation’. With over 10,000+ empowered young people, the Future Perfect program is proof of how technology, when used with care and vision, can make a big difference in society.

This blend of technology, helping people improve their lives, and effective learning, created by The/Nudge and MonsterAPI, not only shows the way to a future where people are skilled in language but also demonstrates the significant change that comes from working together on new ideas. Together, they have not merely created a module; they have fashioned a movement, steering the nation towards a future where ‘language is a bridge to endless opportunities, not a barrier.’

Innovate with a Purpose: Create a Linguistically Inclusive Future

Developers and teams, Join us on this amazing journey! Give MonsterAPI Speech to Text API a try today and see for yourself how it can help overcome language obstacles.

As a special gesture, use the promo code "NUDGE" to get 10,000 FREE API Credits when you join our mission to make the world more inclusive through technology.

The innovation continues, as more and more young voices share stories of change, empowerment, and a future they shape themselves.