update fossasia talk

posts/2023/pycon_india2023_talk/index.qmd

@@ -0,0 +1,37 @@
+---
+title: OpenAI Whisper and it's amazing power to do finetuning
+author: Kurian Benoy
+subtitle: Talk Page for my talk @ Pycon India 2023, Hyderabad
+date: 2023-10-01
+date-format: full
+categories: [conference, fossasia, Audio, Deep learning, SMC, "Talk Page"]
+lang: en
+toc: true
+comments: false
+format:
+  html:
+    highlight-style: a11y
+    theme:
+      light: flatly
+      dark: darkly
+---
+
+## Links to talk
+
+- [Slides](../../../talks/pyconindia-2023/pyconind.qmd)
+
+- Scheduled Time: Sunday, October 1, 2023, 11:00 AM IST 
+
+- [Malayalam ASR Benchmarking project](https://github.com/kurianbenoy/malayalam_asr_benchmarking)
+
+- [Results for Common Voice 11 benchmarking](https://huggingface.co/datasets/kurianbenoy/malayalam_common_voice_benchmarking)
+
+- [Results for SMC MSC benchmarking](https://huggingface.co/datasets/kurianbenoy/malayalam_msc_benchmarking)
+
+## Talk in 30 seconds
+
+I talked about what is OpenAI whisper and some of it's amazing feature. I highlighted the
+importance of fine-tuning Whisper in low-resource language like Malayalam which is able to get
+almost 90% accuracy compared to -10% accuracy of OpenAI whisper model weights. The results where
+calculated with a new github project called [malayalam_asr_benchmarking created by me](https://github.com/kurianbenoy/malayalam_asr_benchmarking).
+