Well Annotated, Gold Standard Training Data to build Multi-lingual Conversational AI
Offered high-quality audio transcription and annotation services, to train their AI-powered Speech Processing Engine.
Project Overview
The Client is India’s pioneering AI product lab with a market presence in the US, India, Canada, and several other countries. Their vision is to augment human potential through technology. They have more than ten years of cutting-edge NLP, scientific research across Artificial Intelligence, ML, Analytics, visualizations, and associated technologies.
They are pushing frontiers in human-machine interaction with their AI products i.e. ICOG and Autovox. ICOG is an AI-based personalized intelligent virtual learning assistant, for workforce learning and transformation goals, whereas, Autovox is an AI-powered Speech
Processing Engine for Indian Languages.
Key Stats
Audio Transcribed & Annotated
1,200 Hrs
No. of
Languages
6 (200 hrs x 6 languages)
Languages Transcribed & Annotated
Indian English, Hindi, Tamil, Telugu, Kannada, Malayalam
Project
Timeline
10 Weeks
Challenges
Lack of access to quality linguists, expert data annotators, and time-to-market have been a bottleneck in the progress & adoption of conversational AI. Sourcing linguists, transcribing and annotating datasets in large volumes with the required quality, sufficient enough to build AI capabilities, has always been a time-consuming, expensive task that requires skilled resources from various domains. Despite having an internal team of annotators, they were facing issues with timely delivery and the quality of annotation.
The critical requirements of the client were:
- Access to Quality Linguists: Lack of access to expert linguists for Indian Languages such as Indian English, Hindi, Tamil, Telugu, Kannada, Malayalam
- Audio Transcription & Annotation: Complex audio annotation & transcription guidelines with minimum accuracy of 95%
- Data Delivery: Transcript files to be delivered in. JSON format within 10 weeks.
Solution
With our deep understanding of conversational AI, we helped the client transcribe and annotate the provided datasets by leveraging a team of expert linguists and annotators to train their AI-powered Speech Processing Engine for Indian Languages.
After evaluating many vendors (including few heavyweights within the industry), The client chooses Shaip because of our expertise to complete large conversational AI projects within stringent timelines, and the quality that we offered at competitive rates.
- Audio Transcription & Annotation Guidelines followed
- Annotated audio type as Speech & Non-Speech
- If the Audio Type of the selected segment is marked as speech, then the type of speech must be selected i.e. Clean Speech, Speech+Music, Speech+Noise, Incomprehensible Speech
- The selected speech segment should be tagged specifically to the language spoken by the speaker
- The annotator should mark the regions with speaker tags. Different speakers should be
marked with different speaker tags. - The annotator must select the gender of the speaker i.e. Male or female.
- The text should be written as spoken in the audio and should be grammatically correct
- If the selected audio type is non-speech, it should be assigned one of the labels – no-speech, music, lip smack, breath, cough, laugh, ring, DTMF, babble, vehicle noise, and intermittent foreground noise
- All the segment in the audio should be marked with tags and then only can be uploaded to the system
- Data Delivery
All transcript files were delivered in JSON format in accordance with the specified metadata requirements within 10 weeks.
The Outcome
The high-quality annotated audio data from expert linguists empowered the client to train their AI-powered speech processing engine accurately in 6 Indian languages i.e. Indian English, Hindi, Tamil, Telugu, Kannada, Malayalam in the stipulated time.
With the gold-standard training datasets, the client will be able to offer an intelligent and robust speech system with multi-lingual capability that uses end-to-end (E2E) and hybrid models to solve real-world problems. In simple terms building Conversational AI (such as Alexa, Siri) specifically catering to the Indian consumers.
We were pleasantly surprised with Shaip’s robust workflow management, quick turnaround time, their experience in conversational AI, and their network of expert linguists. Moreover, the cost value leadership that they offer is second to none.
