Machine Learning from speech to Text - (Speech to text pattern matching)
Budget: $750 – $1,500 USD
This is a 2-stage project:
Stage 1:
From mp3 file with music and speech, use your chosen library such as Deep speech, IBM Watson, Google Speech to Text, Kaldi, Julius, Fairseq, Vosk, Athena, etc. to extract English words.
Store last 30 words in array, but also store every five words or new set every 2 second lack of words into SQLITE db with timestamp, fileID, text.
Compare predefined keywords in SQLITE table (2,000 records) with array and upon any match, store in matches table: fileid, timestamp, keyword
Stage 2:
Replace mp3 with real-time audio via ffmpeg GPIO, sourced from tuner.
Replace fileid with tunerid.
Add custom API trigger to matches table insert (i.e. http://triggerurl.com?tunerid=x&keyw=y
Stage 1:
From mp3 file with music and speech, use your chosen library such as Deep speech, IBM Watson, Google Speech to Text, Kaldi, Julius, Fairseq, Vosk, Athena, etc. to extract English words.
Store last 30 words in array, but also store every five words or new set every 2 second lack of words into SQLITE db with timestamp, fileID, text.
Compare predefined keywords in SQLITE table (2,000 records) with array and upon any match, store in matches table: fileid, timestamp, keyword
Stage 2:
Replace mp3 with real-time audio via ffmpeg GPIO, sourced from tuner.
Replace fileid with tunerid.
Add custom API trigger to matches table insert (i.e. http://triggerurl.com?tunerid=x&keyw=y