The latest articles on DEV Community by Vikas G. (@vikas_g_a248a99bee33e4d3). https://dev.to/vikas_g_a248a99bee33e4d3 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3871581%2F3d65a451-b64a-4c62-b0de-e1b3d3baf6df.png https://dev.to/vikas_g_a248a99bee33e4d3 en Vikas G. Fri, 10 Apr 2026 10:57:15 +0000 https://dev.to/vikas_g_a248a99bee33e4d3/i-built-a-free-call-quality-audit-tool-using-openai-whisper-49a7 https://dev.to/vikas_g_a248a99bee33e4d3/i-built-a-free-call-quality-audit-tool-using-openai-whisper-49a7 <p><strong>TL;DR:</strong> I built <a href="https://www.boloaurlikho.com" rel="noopener noreferrer">boloaurlikho.com</a> — a free tool that transcribes calls and runs quality audits on them. Supports 99+ languages, no signup required. Here's the technical journey and what I learned about building AI-powered QA tools.</p> <h2> The Problem I Was Solving </h2> <p>I run a telecalling operation in India. Our QA process was painful — team leads manually listening to call recordings, filling Excel scorecards, spending 3x the call duration just to audit one conversation. At 500+ calls/day, we were only auditing about 5% of total volume.</p> <p>I wanted something that could:</p> <ul> <li>Transcribe every call automatically (Hindi, English, Hinglish — we use all three)</li> <li>Flag compliance issues without human review</li> <li>Score calls on parameters like greeting, pitch delivery, objection handling, and closure</li> <li>Work without expensive enterprise contracts</li> </ul> <h2> The Tech Stack </h2> <p>The core is OpenAI's Whisper model for speech-to-text. Whisper's multilingual capability was the deciding factor — most transcription APIs choke on code-switched Indian English. Whisper handles "aap ka account number bata dijiye, I'll check the status" without breaking.</p> <p>On top of transcription, I built audit layers:</p> <ul> <li> <strong>Keyword compliance</strong>: Checks if mandatory disclosures, greetings, and CTAs were spoken</li> <li> <strong>Sentiment analysis</strong>: Tracks tone shifts through the conversation</li> <li> <strong>Dead air detection</strong>: Flags excessive silence (usually means agent was scrambling)</li> <li> <strong>Timestamp markers</strong>: So reviewers can jump to problem spots instead of listening to full calls</li> </ul> <h2> What Surprised Me </h2> <p><strong>1. Whisper's accuracy on Indian accents is genuinely impressive.</strong> We tested against Google Speech-to-Text and AWS Transcribe — Whisper won on Hindi and mixed-language content by a significant margin.</p> <p><strong>2. The audit layer is more valuable than the transcription.</strong> Everyone builds transcription tools. The real unlock is what you do with the text after. Automated scoring against custom QA parameters saves 80%+ of manual review time.</p> <p><strong>3. People use it for things I never expected.</strong> Sales teams auditing their own cold calls. Podcast creators checking interview quality. Students verifying lecture transcription accuracy. A lawyer transcribing witness depositions.</p> <h2> Try It </h2> <p>The tool is free at <a href="https://www.boloaurlikho.com" rel="noopener noreferrer">boloaurlikho.com</a>. No signup, no paywall. Supports MP3, WAV, M4A, OGG, WEBM, FLAC. Currently handles files up to 20 minutes, with longer audio support coming soon.</p> <p>If you're building something similar or have questions about the Whisper integration, happy to chat in the comments.</p> ai automation openai showdev