Updates and fixes in 2001/coupard

Format fix README.md.

Update .gitignore with file referenced by the author.

2001/coupard/.gitignore

@@ -1 +1,2 @@
 coupard
+test.wav

2001/coupard/README.md

@@ -1,11 +1,11 @@
 # Most obfuscated sound
 
-    Pierre-Philippe Coupard
+Pierre-Philippe Coupard  
-    Lineo, Inc.
+Lineo, Inc.  
-    ATTN : Pierre-Philippe Coupard
+ATTN : Pierre-Philippe Coupard  
-    390 South 400 West
+390 South 400 West  
-    Lindon, UT 84042
+Lindon, UT 84042  
-    USA
+USA  
 
 ## To build:
 
@@ -27,9 +27,13 @@ make
 ```
 
 If you're using macOS then you won't have these devices. If however you install
-sox from MacPorts or Homebrew you can do something like:
+[sox](https://en.wikipedia.org/wiki/SoX) from
+[MacPorts](https://www.macports.org) or [Homebrew](https://brew.sh) you can do
+something like:
 
-	./coupard | sox -q -traw -r8000 -b8 -e unsigned-integer - -tcoreaudio
+```sh
+./coupard | sox -q -traw -r8000 -b8 -e unsigned-integer - -tcoreaudio
+```
 
 Thanks go to [Yusuke Endoh](/winners.html#Yusuke_Endoh) for providing a proper
 command line (see [/2013/endoh3/README.md](2013/endoh3/README.md).
@@ -45,41 +49,42 @@ your arm or the wall for confirmation.
 Past years' IOCCC entries do many marvellous things in small packages,
 such as console utilities, games, math, languages or even X
 applications.  But they all have one thing in common (despite being hard
-to read obviously) : when they interact with the user, they only
+to read obviously): when they interact with the user, they only
 *display* things, i.e. they display either text or drawings.
 
 So, for a change, this program titillates the sense of hearing instead
-of sight : this program is a self-contained speech synthesizer that
+of sight: this program is a self-contained speech synthesizer that
 tells the current time through the computer's sound card - in just under
 2 Kb of source code !
 
-In the 2001 IOCCC rules and guidelines documents, there is no mention of
+In the 2001 IOCCC [rules](/2001/rules.txt) and
-what is or isn't acceptable regarding audio I/Os, but it is mentioned
+[guidelines](/2001/guidelines.txt) documents, there is no mention of
+what is or isn't acceptable regarding audio I/O, but it is mentioned
 that hardware-specific and/or non-portable programs are not welcome. So,
 in order to do audio and satisfy those requirements, the format required
-by /dev/audio has been chosen : /dev/audio is a rather ancient
+by `/dev/audio` has been chosen: `/dev/audio` is a rather ancient
 fixed-format audio interface that is supported by the majority of Unices
-that can do sound, as a lowest-denominator sound I/Os interface.
+that can do sound, as a lowest-denominator sound I/O interface.
 
-### Program usage:
+### Program usage
 
-* To listen to what the program says directly, invoke
+To listen to what the program says directly, invoke
 
 ```sh
 ./coupard > /dev/audio
 ```
 
-the program then speaks out something like
+The program then speaks out something like
 
   "The time is HH hours MM minutes SS seconds"
 
 in military time (24 hours). Of course, you need to have permissions
-to write to /dev/audio, have your sound card enabled, the volume up,
+to write to `/dev/audio`, have your sound card enabled, the volume up,
 the speakers connected ... :-)
 
-* If you want to convert the audio output of the program into another
+If you want to convert the audio output of the program into another audio
-audio format, you can pipe it to sox for example (sox is the
+format, you can pipe it to [sox](https://en.wikipedia.org/wiki/SoX) for example
-Swiss army knife of Unix audio tools) : the command
+(`sox` is the Swiss army knife of Unix audio tools): the command:
 
 ```sh
 ./coupard | sox -c1 -r8000 -tub - -c2 -r44100 -twav test.wav
@@ -90,28 +95,30 @@ file and saves it into "test.wav".
 
 The output format of the program is 8KHz, unsigned 8-bit samples.
 
-NOTE : the program's return code is meaningless
+NOTE: the program's return code is meaningless.
 
-### Audio quality:
+### Audio quality
 
 The quality of the sound produced by the program is cross between a
-yogurt-pot-and-string telephone and a bad gramophone, i.e. the S/N
+yogurt-pot-and-string telephone and a bad gramophone, i.e. the S/N ratio is very
-ratio is very low and the signal's top frequency is very low as well.
+low and the signal's top frequency is very low as well.  This is because the
-This is because the audio is stored internally at a sampling rate of
+audio is stored internally at a sampling rate of 4KHz (the [Nyquist
-4KHz (the Nyquist frequency is only 2KHz) with only 4-bit samples
+frequency](https://en.wikipedia.org/wiki/Nyquist_frequency) is only 2KHz) with
-(adding 24db of quantization noise to an already poor signal).
+only 4-bit samples (adding 24db of quantization noise to an already poor
+signal).
 
 In short, you might have to listen twice to understand what the
 program is saying. 2048 bytes including the sound generation program
 doesn't leave much room for audio quality ;-)
 
-### Speech generation:
+### Speech generation
 
 Only 9 phonemes are used to generate the speech. Beside the limited
 number of words the program has to generate (66 words), the low sound
 quality is actually exploited to take liberties with certain
 consonants (for example, "s" and "f" sound very much the same at 4KHz,
-so they can be merged into one phoneme). Because the psychoacoustic
+so they can be merged into one phoneme). Because the
+[psychoacoustic](https://en.wikipedia.org/wiki/Psychoacoustics)
 context around those phonemes is just sufficient, the words can be
 interpreted correctly most of the time despite the fact that they are
 pronounced quite wrongly.
@@ -120,7 +127,7 @@ Internally, a table is used to generate most of the words, and
 composite words such as "thirty-five" are made up by the C code
 externally.
 
-### Data compression:
+### Data compression
 
 Even at 4KHz / 4 bits, the weight of all the phoneme samples and the
 word-to-phonemes table put together is over 1600 bytes, which is well
@@ -134,22 +141,21 @@ decompression code can't be made small enough.
 
 The best compression compromise was found using a simple Huffman
 compressor, and encoding the compressed data by blocks of 13 bits in
-pairs of printable ASCII characters : the raw data is compressed to
+pairs of printable ASCII characters: the raw data is compressed to
 8465 bits, and the resulting C string is therefore about 1300
 characters, which leaves just enough room for the decompressor with a
 hard-coded Huffman tree, and the code that plays the phonemes in the
 right order.
 
-### Note on compiler warnings and lint:
+### Note on compiler warnings and lint
 
-The program should build with gcc's -Wall -Werror and -ansi flags.  If
+The program should build with `gcc -Wall -Werror -ansi`.  If the `-pedantic`
-the -pedantic flag is used, gcc will complain about the length of the
+flag is used, gcc will complain about the length of the data string and the sign
-data string and the sign of its type, and also the fact that main()
+of its type, and also the fact that `main()` returns with no value. The
-returns with no value. The confusion with the sign of the char type
+confusion with the sign of the char type doesn't matter with the data string
-doesn't matter with the data string because it's composed only of
+because it's composed only of printable characters under 127.
-printable characters under 127.
 
-When linting the program with lclint, it complains about a huge list
+When linting the program with `lclint`, it complains about a huge list
 of things that, if corrected to make lint happy, would probably double
 the size of the source code.