Sentiment Call Analyzer
Automated call transcription and sentiment analysis using Google Cloud Speech-to-Text and BigQuery, improving insight delivery speed by 25%.
Software Engineer · January 2021 – August 2025
Software Engineer
DataGrokr Analytics
Python · Google Cloud Speech-to-Text · BigQuery · GCP · SQL
01 · Context
Overview
An automated pipeline that ingests call recordings, transcribes them using Google Cloud Speech-to-Text, runs sentiment analysis over the resulting transcript, and stores structured results in BigQuery for analysis and reporting.
Business Problem
Call recordings were reviewed manually to extract sentiment signals and identify key topics — a time-intensive process that capped the volume of calls that could be analysed and delayed delivery of insights. Teams could only review a sample of recordings, limiting visibility of patterns across the full call volume.
My contribution
Designed and implemented the full pipeline: Speech-to-Text integration, transcript processing, sentiment analysis, BigQuery data modelling, and the reporting layer.
02 · System
Solution Approach
An automated workflow that ingests call recordings, transcribes audio to text using Google Cloud Speech-to-Text, runs sentiment analysis over the transcript to extract sentiment signals and key topics, normalises results into structured records with call metadata, and writes them to BigQuery for querying and reporting.
Workflow
Call Audio
Call recordings ingested from telephony system or cloud storage.
Speech-to-Text
Google Cloud Speech-to-Text converts audio to a time-stamped text transcript.
Transcript Processing
Transcript cleaned and segmented for downstream analysis.
Sentiment Analysis
Sentiment signals and key topics extracted from the processed transcript.
BigQuery Storage
Results normalised into structured records and written to BigQuery with call metadata.
Actionable Insights
Analysts query structured results for trend analysis, reporting, and business decision support.
03 · Engineering
Technical Decisions
Google Cloud Speech-to-Text used for transcription — provided reliable audio-to-text conversion without requiring a custom speech model, and integrated naturally with the GCP stack.
BigQuery used as the analytics store — enabled SQL-based analysis at scale without additional data warehouse infrastructure, accessible to analysts already familiar with the environment.
Structured data modelling designed to support both per-call queries and aggregate trend analysis across cohorts of calls over time.
Engineering Challenges
Variable audio quality in real call recordings — background noise, overlapping speech, and connection artefacts affected transcription accuracy and required resilient handling of partial transcription output.
Designing sentiment analysis to distinguish topic sentiment from call sentiment — discussions of negative topics do not necessarily indicate negative customer sentiment.
Ensuring analysis results were structured consistently enough to support aggregate querying across large call volumes while preserving per-call detail.
04 · Outcome
Verified Results
25% improvement in insight delivery speed compared to the previous manual review process
Ability to analyse the full call volume rather than a sampled subset — surfacing patterns that spot-checks could not reliably detect
Structured, queryable sentiment data in BigQuery enabling trend analysis across call cohorts
Automating speech-to-insight pipelines delivers compounding value: the immediate gain is speed, but the structural gain is coverage — analysing 100% of call volume rather than a sample reveals patterns that selective manual review consistently misses.
Get in touch
shalinirai450@gmail.com