Split one WAV file into many clips of 30sec duration

api_batch.rb

#!/usr/bin/env ruby
# Submit all WAV clips to AT&T Speech API

require 'thread'
require 'rest-client'
require 'json'

src_hash = ARGV[0]
index = ARGV[1].to_i
bearer_token = ARGV[2]

url = "https://api.att.com/speech/v3/speechToText"

puts "using hash #{src_hash}, count #{index}"

class Worker
  
  def initialize(n,hash,url,token)
    @index = n
    @hash = hash
    @url = url
    @token = token
  end
  
  def run
    file_name = "#{@hash}-#{"%02d" % @index}"
    if FileTest.exists?("#{file_name}.json")
      puts "skipping segment #{@index}"
      return
    else
      puts "processing segment #{@index}"
    end
    
    res = RestClient::Resource.new(@url)
    rsp = res.post File.open("#{file_name}.wav", "r"), 
      "Accept" => "application/json",
      "Content-Type" => "audio/wav",
      "Authorization" => "Bearer #{@token}"
    if rsp.body.length>0
      file = File.open("#{file_name}.json", "w")
      file.write(rsp.body)
      file.close
    else
      puts "API response empty for segment #{@index}"
    end
    
  rescue => e
    puts "error encountered: #{e}"
    puts rsp.body
  end
  
end

queue, processed = Queue.new, Queue.new

for n in (1..index)
  queue << Worker.new(n, src_hash, url, bearer_token)
end

10.times do
  Thread.new { processed << queue.pop.run until queue.empty? }
end

while processed.length<index do
  sleep 1
end

speech_post.rb

#!/usr/bin/ruby
# Ruby JSON processor for results from AT&T Speech API

require 'json'

src_hash = ARGV[0]
index = ARGV[1].to_i

all_words = []
all_scores = []

index.times do |n|
	file = File.open("#{src_hash}-#{"%02d" % (n+1)}.json")
	json = JSON.parse(file.read)
	
	nbest = json['Recognition']['NBest']
	if nbest
		words = nbest[0]['Words']
		scores = nbest[0]['WordScores']
		
		all_words = all_words + words.map {|e| e.gsub(/\./,'')}
		all_scores = all_scores + scores
	end
end

puts "transcript:"
puts all_words.join(" ")

threshold = 0.5
meaningful_words = all_words.select.with_index {|e,k| all_scores[k]>threshold} - ['the', 'of', 'with', 'for', 'that', 'you', 'and', 'in', 'to', 'a', 'about', 'is', 'I', 'on', 'be', 'like', 'as', 'this', 'but', 'we', 'really', 'have', 'know', 'all', 'if', 'can', 'not', 'they', 'are', 'very', 'because', 'it', 'do', 'them', 'get', 'more', 'will', 'your', 'so', 'how', 'one', 'at', 'something', 'right']

puts "\n\nfound #{meaningful_words.uniq.count} unique meaningful words"

puts "top 25:"
raw_counts = Hash[meaningful_words.uniq.map {|e| [e,meaningful_words.select {|n| n==e}.count]}]
top_words = meaningful_words.uniq.sort {|a,b| 0.75 * raw_counts[b] + 0.75 * b.length <=> 0.75 * raw_counts[a] +  0.75 * a.length}.take(25)
puts top_words.map {|e| "#{raw_counts[e]} #{e}"}

wav_split.rb

#!/usr/bin/ruby
# Process a WAV file into short clips

require 'wavefile'
require 'digest'
include WaveFile

# read input file and deconstruct into 30sec chunks
src_file_name = ARGV[0]
src_hash = Digest::SHA1.hexdigest(src_file_name)[0..9]

reader = Reader.new(src_file_name)
writer = nil
index = 1
begin
  total = 0
  writer = Writer.new("#{src_hash}-#{"%02d" % index}.wav", Format.new(:mono, :pcm_16, 32000))
  while true do
    buffer = reader.read(4096)
    writer.write(buffer)
    total = total + buffer.samples.count
    if total >= 30 * 32000
      total = 0
      index = index + 1
      writer.close
      putc "."
      writer = Writer.new("#{src_hash}-#{"%02d" % index}.wav", Format.new(:mono, :pcm_16, 32000))
    end
  end
rescue EOFError
  reader.close
  writer.close 
end
puts

puts "using src hash #{src_hash}, split into #{index} files"