Speech To Text: Turning Audio into Text

#The problem

Listening to long recordings and typing them out by hand took hours. Meeting, interview and call recordings needed to become text quickly, and be searchable.

#What I did

• Built a system that takes audio files and live speech and turns them into text.
• Pre-processed audio to improve accuracy on Turkish speech: noise reduction, level balancing, splitting long recordings into chunks.
• Made the output readable with punctuation and paragraphs, and added timestamps to jump to the right moment in a recording.
• Stored transcripts so they're searchable: search a word and see which recording and which minute it appears in.
• Exposed the system as an API that existing applications can connect to.

#The result

Recordings become readable text within minutes. Manual transcription is largely gone, and finding information in old recordings is easy.