Manas Mallik

Part of team researching core algorithm (Vector Quantization, HMM etc) for reliable
speaker recognition and verification.
Develop IVR for training and testing the system
Releasing the system as API for integrating with different products (internal and
external)
Speaker recognition and verification is a system wherein users calling into the system can be authenticated. The technology has a high number of use cases in production systems, for example banks can use… Show more

Part of team researching core algorithm (Vector Quantization, HMM etc) for reliable
speaker recognition and verification.
Develop IVR for training and testing the system
Releasing the system as API for integrating with different products (internal and
external)
Speaker recognition and verification is a system wherein users calling into the system can be authenticated. The technology has a high number of use cases in production systems, for example banks can use it as a preliminary screening method for verifying if the calling user is really the one he/she claims to be. Show less

VAS- IVR Learning Application for Vocal Music with auto-feedback system. Front end is a full fledged ivr with different chapters, exercises, tests, feed-backs. Backend is complete Vocal Analyzer which analyses a vocal recording according to various musical evaluation parameters.
Language Used: Java

Responsibilities:
Developing the core algorithm behind the Singing performance rating engine (The system evaluates the user singing performance on various music parameters like, scale, melody, tempo etc)

Karaoke is a Music Rating System used by the operator’s subscribers, wherein users choose and sing songs from a given list, and the system rates their singing based on different parameters.
It reuses most of call handling, alert, reporting, analytics and… Show more

Responsibilities:
Developing the core algorithm behind the Singing performance rating engine (The system evaluates the user singing performance on various music parameters like, scale, melody, tempo etc)

Karaoke is a Music Rating System used by the operator’s subscribers, wherein users choose and sing songs from a given list, and the system rates their singing based on different parameters.
It reuses most of call handling, alert, reporting, analytics and integration system of Music Search product. Show less

Saptak Music Player: Is a cloud based Music Player, where you can search any song, song gets downloaded and played simultaneously, User Define playlist stored in cloud.
Module: Encryption Algorithm for the media files, so that they can be played using any other player, Ogg Player ( Decrypts and plays Encrypted Ogg Files)
Module: Local Media Manager, detect the local songs, tags them if required using Audio Finger Printing
Language: C++ with Qt

A music companion, which would help a vocalist to practice singing. A combination of tabla/tanpura (where tals and sequence of tones can be selected) and midi instruments and a note editor(notes can be written which can be played in loop)
Language Used: Java

Design & Develop an Vocal Singing Evaluation Algorithm for a providing feedback to singing in a Karaoke Application application to be deployed as a VAS service, by the network/telecom operators.
Language Used: Java

Song Catcher Server - Using Indexing Algorithm: Index(Subfingerprint:Postion) based on fact that for small distortion Bit Error Rate for Fingerprint is less than 2 the search is reduced to matching only positions corresponding the to Subfingerprint as opposed to brute force, using the GPU brute force server as a fallback option
Language: C, Java

Design & Develop a high throughput Song Catcher Server, using GPUs(NVidia GForce Cards) for parallel processing the brute force matching required to identify the song from a database of 2 million songs
Language: C / Cuda APIs

My roles include:
Research & Developing the core algorithms for music recognition(identification of Song or music piece from a recorded/gsm distorted small portion of the song).
Designing Efficient/Scalable Architecture for handling a huge number of songs

Music Discovery system is a unified service comprising the following customer facing
modules:
--Call Handling System: Frontend system to handle user calls/requests. It is an extremely flexible system and… Show more

My roles include:
Research & Developing the core algorithms for music recognition(identification of Song or music piece from a recorded/gsm distorted small portion of the song).
Designing Efficient/Scalable Architecture for handling a huge number of songs

Music Discovery system is a unified service comprising the following customer facing
modules:
--Call Handling System: Frontend system to handle user calls/requests. It is an extremely flexible system and can be tuned to changing customer requirements in order of minutes.
--SongCatcher: Catches music playing on audio source like TV/Radio
--GanaBolo: Recognizes spoken keywords to identify the song user wishes to search
--Text Search: Searches written keywords from a vast metadata of songs. Search can get results based on title, album, artistes, actors etc.
--Recommendation Engine: Gets recommended songs for a user based on past history, user similarity and content similarity.
Following are the some of the other supporting modules:
--Alert System: Monitors system health and, based on severity, alerts concerned people in case any anomalies are observed. Alerts can come in form of email, SMS, IM ping and Call.
--Reporting system and MIS: A full-fledged reporting system for sending internal and external reports.
--Analytics System: Uses different business rules to extract data from a huge repository for business intelligence and future decisions.
--Integration System: Exposes different modules through Web Services for third parties.
Other systems like Promotion Manager, Dashboards etc. are used to increase revenue and give better insight into the system Show less

Design & Develop an Audio Fingerprinting Algorithm for a Song Identification( Music Recognition) application to be deployed as a VAS service, and to be integrated with an Ring Back Tone(HelloTune) setting service by the network/telecom operators.
Language Used: C

Design & Develop voice tagging application for a voice dialing vas service having the constraint of having training data restricted to very few samples.
Languages & Tools: Perl, HTK, C

Design & Develop Indian Accented English Text-to-Speech Synthesis Engine.
Language Used: C++
Role / Responsibility: Lead, Research, Design & Develop

Design & Develop tools for voice generations for TTS Engine
Language Used: C++, Shell( awk sed) scripts
Role / Responsibility: Lead & Develop

Would be used to connect to the remote product (Avaya IR ) boxes and do all the thing required to enable them into Enterprise Services Platform of Avaya for providing Managed services
Languages Used: Expect/tcl/tk, shell, awk, sed scripts
Role / Responsibility: Product Study & Develop

It is a component which can behave like modem for various avaya Product which use dial out alarming mechanism, capture the alarms and send them as SNMP traps
Languages Used: C, Expect/tcl, shell scripts
Role / Responsibility: Product Study & Develop

Mobile Payment system is meant to enable payments to happen by bringing two phones close to each other and using the Data Tramission algorithm via Audio Transmission, without requiring the need of internet on phones. The payment solution is such that transaction can happen using bank a/c , credit & debit cards, system having a following components

- Customer Application - Android application where customer can link his bank accounts/ credit/debit cards, and once done can… Show more

Mobile Payment system is meant to enable payments to happen by bringing two phones close to each other and using the Data Tramission algorithm via Audio Transmission, without requiring the need of internet on phones. The payment solution is such that transaction can happen using bank a/c , credit & debit cards, system having a following components

- Customer Application - Android application where customer can link his bank accounts/ credit/debit cards, and once done can use it for payment without requiring any internet & recharges over internet or remote payments to the merchant
- Merchant Application – Android application on which merchant can receive payments, check his transaction history, or check remote payments
- Customer Application Server - Which will talk to customer application for registration , bank addition reward point updates, transation status, notifications.
- Merchant Application Server – Which will talk to the merchant application for login, transaction and heartbeat of the merchant app.
The system is complex, due to ensuring reliable data transfer using sound waves, proper encryption and decryption and key exchanges between backend server and user’s clients, android application development, and finally, integration between backend servers and APIs of different banks and of NPCI and Payment Gateways Show less

Developed a protocol stack for transferring data between two devices over high frequency sound, this can used as Near Field communication without any additional hardware on mobile phones.
Proof of Concept/Demo Android Application showing data transfers based on above technology.
- Payment App PoC
- Contact Transfer PoC
- Music Transfer PoC

Vaachak is an Indian Language Text-to-Speech (TTS) Engine. The TTS system consists of text pre-processor (TPP) module, which takes the raw text as an input and tokenizes the text. Text-to-phoneme (sound units) conversion is the Dictionary-based as well as Letter to Sound (LTS) Rule-based approach, in which pronunciations rules are applied to words based on their spellings. These sound units are concatenated using voice corpus to generate synthesized speech.

Development of RPC Based server and Client APIs for TTS.
Language/Technology Used: C++, RPC

Implemented Phone based recognizer system using HTK(Hidden markov model toolkit) for automatic labeling/alignment of recorded speech to the corresponding phonetic sequence.
Language/Tool Used: Perl

Prototyping and Implementation a module for Analysis and Re-Synthesis of Speech Signal using Harmonic plus Noise modeling.
Language/Tool Used: Matlab, C++

Implementation PSOLA (Pitch Synchronous Overlap and Addition) technique for modification of time and pitch of recorded speech.
Language Used: C

Prototyped a module for reconstruction of Speech from mel-frequency cepstral coefficients and pitch frequency in Matlab
Language/Tool Used: Matlab, C++

Speech Synthesis Engine using Data Driven or Corpus Based approach.
--Automatically clustered similar unit using tree based clustering.
--Implemented unit Selection using viterbi decoding.
Process and tools for Speech Corpus Preparation for Data Driven synthesis
Language: C++, CMU Speech Tools, HTK

Framework for building Text-to-Speech system for Indian Languages
--Incorporated rule Interpreter for LTS and syllabification rules for Indian language.
--Developed LTS rules for hindi and Indian English as reference implementation.
--Developed TTS voices for hindi, Indian English and Indian Names
Enhance Data Driven Synthesis Engine to synthesize using non-uniform units (phoneme, diphone, syllable) to improve quality of speech
Reduction of footprint of Data driven Text to speech… Show more

Framework for building Text-to-Speech system for Indian Languages
--Incorporated rule Interpreter for LTS and syllabification rules for Indian language.
--Developed LTS rules for hindi and Indian English as reference implementation.
--Developed TTS voices for hindi, Indian English and Indian Names
Enhance Data Driven Synthesis Engine to synthesize using non-uniform units (phoneme, diphone, syllable) to improve quality of speech
Reduction of footprint of Data driven Text to speech system, to 20-25mb and improvement of TTS performance to around 100ms for an average length sentence.
Project Lead, enhancement of quality of Output speech for Data Driven Synthesis using:
--Consideration of word and word aggregate level unit
--Treating word boundary unit as separate units.
--Improved cost function to include pitch carry forwarding for unvoiced units.

Show less

1. Architecture of a basic Text-to-Speech System
2. LTS (letter-to-sound) module for phonetization of hindi text.
3. Speech Synthesis Engine using formant Based Approach using Klatt’s Formant Synthesizer
4. Prosody Model for (Imparting Naturalness to Synthesized Speech)
- Linguistic Analysis of hindi text for evaluation of context.
- Model for application of Intonation, stress, fluency to synthesized speech
5. Speech Synthesis Engine using… Show more

1. Architecture of a basic Text-to-Speech System
2. LTS (letter-to-sound) module for phonetization of hindi text.
3. Speech Synthesis Engine using formant Based Approach using Klatt’s Formant Synthesizer
4. Prosody Model for (Imparting Naturalness to Synthesized Speech)
- Linguistic Analysis of hindi text for evaluation of context.
- Model for application of Intonation, stress, fluency to synthesized speech
5. Speech Synthesis Engine using Concatinative (concatenation of recorded speech units) approach.
Show less

Project Lead, Developed an Application to Demonstrate TTS product having IVR, Web and SMS Interfaces.
Language Used:C++, JavaScript, VXML, AT command