Improve WAV parser (#243)

* Skip parsing Fact chunks for WAV parser

* Improve inline documentation

* Small improvements

* Be more lenient when returning available metadata

* Reverse test and mention regression

* Put back the fact logic

* Update test

* Bump version

* Refactor extracting fmt data

Author: linkyndy

CHANGELOG.md

@@ -1,5 +1,8 @@
+## 2.9.0
+* Improve WAV parser by performing a best-effort when extracting metadata from files that do not strictly follow the format spec.
+
 ## 2.8.0
-* Add support for Ruby 3.2 and 3.3
+* Add support for Ruby 3.2 and 3.3.
 
 ## 2.7.2
 * Improved stability for mp4 parser when dealing with corrupted FTYP boxes.

lib/format_parser/version.rb

@@ -1,3 +1,3 @@
 module FormatParser
-  VERSION = '2.8.0'
+  VERSION = '2.9.0'
 end

lib/parsers/wav_parser.rb

@@ -20,75 +20,60 @@ def call(io)
     # with the exception that the Format chunk must precede the Data chunk.
     # The specification does not require the Format chunk to be the first chunk
     # after the RIFF header.
-    # http://soundfile.sapp.org/doc/WaveFormat/
-    # For WAVE files containing PCM audio format we parse the 'fmt ' and
-    # 'data' chunks while for non PCM audio formats the 'fmt ' and 'fact'
-    # chunks. In the latter case the order fo appearence of the chunks is
-    # arbitrary.
-    fmt_processed = false
-    fact_processed = false
+    # https://www.mmsp.ece.mcgill.ca/Documents/AudioFormats/WAVE/WAVE.html
     fmt_data = {}
-    total_sample_frames = 0
+    data_size = 0
+    total_sample_frames = nil
     loop do
       chunk_type, chunk_size = safe_read(io, 8).unpack('a4l')
       case chunk_type
       when 'fmt ' # watch out: the chunk ID of the format chunk ends with a space
         fmt_data = unpack_fmt_chunk(io, chunk_size)
-        return process_non_pcm(fmt_data, total_sample_frames) if fmt_data[:audio_format] != 1 and fact_processed
-        fmt_processed = true
       when 'data'
-        return unless fmt_processed # the 'data' chunk cannot preceed the 'fmt ' chunk
-        return process_pcm(fmt_data, chunk_size) if fmt_data[:audio_format] == 1
-        safe_skip(io, chunk_size)
+        data_size = chunk_size
       when 'fact'
         total_sample_frames = safe_read(io, 4).unpack('l').first
         safe_skip(io, chunk_size - 4)
-        return process_non_pcm(fmt_data, total_sample_frames) if fmt_processed and fmt_data[:audio_format] != 1
-        fact_processed = true
       else
         # Skip this chunk until a known chunk is encountered
         safe_skip(io, chunk_size)
       end
+    rescue FormatParser::IOUtils::InvalidRead
+      # We've reached EOF, so it's time to make the most out of the metadata we
+      # managed to parse
+      break
     end
+
+    file_info(fmt_data, data_size, total_sample_frames)
   end
 
   def unpack_fmt_chunk(io, chunk_size)
     # The size of the fmt chunk is at least 16 bytes. If the format tag's value is not
     # 1 compression might be in use for storing the data
     # and the fmt chunk might contain extra fields appended to it.
-    # The last 4 fields of the fmt tag are always:
+    # The first 6 fields of the fmt tag are always:
+    # * unsigned short     audio format
     # * unsigned short     channels
     # * unsigned long      samples per sec
     # * unsigned long      average bytes per sec
     # * unsigned short     block align
     # * unsigned short     bits per sample
 
-    fmt_info = safe_read(io, 16).unpack('S_2I2S_2')
+    _, channels, sample_rate, byte_rate, _, bits_per_sample = safe_read(io, 16).unpack('S_2I2S_2')
     safe_skip(io, chunk_size - 16) # skip the extra fields
 
     {
-      audio_format: fmt_info[0],
-      channels: fmt_info[1],
-      sample_rate: fmt_info[2],
-      byte_rate: fmt_info[3],
-      block_align: fmt_info[4],
-      bits_per_sample: fmt_info[5],
+      channels: channels,
+      sample_rate: sample_rate,
+      byte_rate: byte_rate,
+      bits_per_sample: bits_per_sample,
     }
   end
 
-  def process_pcm(fmt_data, data_size)
-    return unless fmt_data[:channels] > 0 and fmt_data[:bits_per_sample] > 0
-    sample_frames = data_size / (fmt_data[:channels] * fmt_data[:bits_per_sample] / 8)
-    file_info(fmt_data, sample_frames)
-  end
-
-  def process_non_pcm(fmt_data, total_sample_frames)
-    file_info(fmt_data, total_sample_frames)
-  end
-
-  def file_info(fmt_data, sample_frames)
-    return unless fmt_data[:sample_rate] > 0
-    duration_in_seconds = sample_frames / fmt_data[:sample_rate].to_f
+  def file_info(fmt_data, data_size, sample_frames)
+    # NOTE: Each sample includes information for each channel
+    sample_frames ||= data_size / (fmt_data[:channels] * fmt_data[:bits_per_sample] / 8) if fmt_data[:channels] > 0 && fmt_data[:bits_per_sample] > 0
+    duration_in_seconds = sample_frames / fmt_data[:sample_rate].to_f if fmt_data[:sample_rate] > 0
     FormatParser::Audio.new(
       format: :wav,
       num_audio_channels: fmt_data[:channels],

…/fixtures/WAV/invalid_d_6_Channel_ID.wav spec/fixtures/WAV/d_6_Channel_ID.wavspec/fixtures/WAV/invalid_d_6_Channel_ID.wav renamed to spec/fixtures/WAV/d_6_Channel_ID.wav spec/fixtures/WAV/invalid_d_6_Channel_ID.wav renamed to spec/fixtures/WAV/d_6_Channel_ID.wav

@@ -46,9 +46,14 @@
     expect(parse_result.media_duration_seconds).to be_within(0.01).of(13.81)
   end
 
-  it "cannot parse file with audio format different from 1 and no 'fact' chunk" do
-    expect {
-      subject.call(File.open(__dir__ + '/../fixtures/WAV/invalid_d_6_Channel_ID.wav', 'rb'))
-    }.to raise_error(FormatParser::IOUtils::InvalidRead)
+  it 'returns correct info about non pcm files with no fact chunk' do
+    parse_result = subject.call(File.open(__dir__ + '/../fixtures/WAV/d_6_Channel_ID.wav', 'rb'))
+
+    expect(parse_result.nature).to eq(:audio)
+    expect(parse_result.format).to eq(:wav)
+    expect(parse_result.num_audio_channels).to eq(6)
+    expect(parse_result.audio_sample_rate_hz).to eq(44100)
+    expect(parse_result.media_duration_frames).to eq(257411)
+    expect(parse_result.media_duration_seconds).to be_within(0.01).of(5.83)
   end
 end