Description:

Basic Summary:

Voice automation heavily relies on activation phrases such as "Ok google", "hey alexa" and "Hey siri" to start listening to user. This is a first of a kind foss system which allows anyone to create their own hotword detector with very less data

Tentative agenda :

Introduction (5 minutes)

• Briefly introduce the topic of hotword detection and its importance in voice assistants and other applications.
• Explain the limitations of traditional machine learning approaches for hotword detection and the need for one-shot learning.
• Introduce EfficientWord-Net and its key features.

History of Voice Assistants (10 minutes)

• Use the Voice Assistant Timeline[1] to provide a brief history of voice assistants from the 1960s to the present day.
• Highlight the key milestones and developments in voice assistant technology, including the rise of smart speakers and voice-activated devices.
• Discuss the challenges and opportunities presented by the proliferation of voice assistants.

Hotword Detection Techniques (10 minutes)

• Explain the different techniques used for hotword detection, including keyword spotting, neural network-based approaches, and one-shot learning.
• Discuss the advantages and disadvantages of each technique and their suitability for different applications.
• Focus on one-shot learning and its benefits for hotword detection.

EfficientWord-Net Architecture (10 minutes)

• Provide an overview of the EfficientWord-Net architecture, including its use of a Siamese network and Tflite implementation.
• Explain how EfficientWord-Net performs hotword detection using only a few training samples.
• Discuss the advantages of EfficientWord-Net over traditional machine learning approaches for hotword detection.

Demo and Use Cases (10 minutes)

• Provide a live demo of EfficientWord-Net in action, using a Raspberry Pi or other device.
• Discuss the use cases for EfficientWord-Net, including voice assistants, home automation, and other applications.
• Highlight the benefits of using EfficientWord-Net for custom hotword detection and its potential impact on the voice assistant market.

Conclusion and Q&A (5 minutes)

• Summarize the key points of the talk and highlight the importance of EfficientWord-Net for hotword detection.
• Open up the floor for questions and discussion.

Note: The timings are approximate and can be adjusted based on the audience and the level of detail required

Prerequisites:

knowledge in python and home automation

Video URL:

https://drive.google.com/file/d/1aRncUnq-P_iddU-8H0C4qrV_fmrxZGGx/view?usp=sharing

Content URLs:

https://github.com/Ant-Brain/EfficientWord-Net

Speaker Info:

I am a computer science graduate currently working as the lead AI engineer in Trademarkia ( a legal tech law firm). I have over 2 years of experience in AI. I am tinkerer by nature so have a basic yet functional knowledge in ECE as well​

Speaker Links:

https://www.linkedin.com/in/chidha1434/ https://github.com/Ant-Brain/EfficientWord-Net