Using a Digital Voice Recorder for Speech Recognition, Part II

Continuing yesterday's exhaustive discussion of the Sony ICD-MX20 recorder I'm using for input to my speech-recognition software, it may be worth clarifying my use of the term high fidelity in the decidedly lo-fi context of human speech. In the larger realm of hi-fi audio, this recorder's 60-13,500 Hz frequency response is laughable; it's a far cry from the 20-20,000 Hz response that defines the minimum qualifications of any recording or playback equipment suitable for the world of music. But when it comes to capturing the human voice for purposes such as speech recognition, this recorder is more than adequate. Of course, frequency response is only one element in the equation; things like sampling rate and the absence of various types of noise also affect the quality of any recording. In any event, my use of the hi-fi terminology in this context is purely relative. It's intended to draw a distinction between using this particular recorder for speech recognition, versus other digital voice recorders that aren't necessarily up to the task.

Like other digital voice recorders offered by various manufacturers, this one uses a proprietary file scheme. Sony's LPEC format results in efficient use of memory, or disk space when the file is transferred to your computer. This is especially important because of this unit's stereo-recording capabilities; two channels are better than one, but also require about twice the space of the monophonic equivalent. Although the MX20 has no native ability to record or play MP3-format material, the included Windows software will convert an MP3 track to the required LPEC format. A 5.06 MB MP3 track I converted in this way resulted in a 1.95 MB LPEC file; the MP3 file had been encoded at a similar bit-rate as that used by the MX20 at its highest-quality setting. Audio quality was noticeably reduced: the high end of the audio spectrum fell into the background, and a subwoofer would have had nothing to do, had this been played back through a hi-fi audio system. It isn't rotten, but you'd miss your brass percussion and anything else that normally lives in the higher treble regions, including, of course, those precious harmonic frequencies that make all the difference between listening excitement and lethargy. Many Podcasts, on the other hand, would be perfectly acceptable after such a conversion.

In addition to a number of other conversions to various Sony formats—useful, I suppose, if you have other Sony-branded audio equipment in your stable—the included software also allows you to save your recording as a .wav file. I wasn't able to reverse the process—WAV to LPEC—so this type of conversion would rely on a three-step WAV-MP3-LPEC procedure, unless there's a Sony CODEC available for this purpose. I'm not aware of one, but that doesn't necessarily mean that one doesn't exist. The included software, according to the documentation, doesn't run on Windows 95, 98, or NT platforms, nor under Macintosh operating systems. Connection to your computer is via an included USB cable. The recorder appears, to Windows, as one or two disk drives: one if the internal memory is used by itself, and the second if a Memory Stick is installed. As you might expect, you can use the memory for storing other types of files as well.

Used as a stereo recorder, the unit excels in its ability to capture conversations that sound virtually identical to the way you heard them yourself, with your own ears. Even when the recorder is in a shirt pocket—and barring extraneous noises generated by bodily movement and rustling garments—the result is virtually identical to what actually occurred. I've made a few test recordings of this nature, and excluding extremely difficult acoustics in very large rooms, I had no trouble picking out individual voices during playback.

Like your computer, the MX20 organizes recordings using the familiar folders metaphor. Regardless of memory capacity, the maximum number of folders is 340 apiece for both the built-in memory and any installed Memory Stick. The maximum number of recordings per folder is 999.

It's possible, using just the recorder by itself, to divide a recording into two parts during recording or playback. It's also possible to add to the end of an existing recording, or overwrite a portion of an existing recording from a particular point. Using the included software, you can also combine recordings. There's no provision for recording while the MX20 is connected to a computer via USB.

Recordings can be assigned a priority, from one to three, which determines how they're sorted. The default sorting scheme is by recorded date and time. Recordings can be moved from one folder to another, and can be erased one at a time or all at once.

You can assign an alarm to a recording. When the alarm time arrives, the unit will beep, or beep and then play back a recording of your choosing. This is useful if you have a hard time waking up to a normal alarm sound; a recording of your mother shrieking at you to get out of bed may be more effective. My personal favorite is a recording of a tribe of enraged baboons, but the possibilities are endless.

Menu operations are via a joystick located between the record and stop buttons on the front panel. On the rear is a switch that's simply labeled hold. During recording or playback the switch functions in the expected way: operating controls are disabled to prevent accidents. When the MX20 isn't being used for recording or playback, the same switch puts the unit into a kind of quiescent state. There's no power switch on this unit, resulting, probably, in a decrease in battery life but also an increase in usability. One benefit of doing things this way is that the recorder is always ready to go; there's no delay while the unit powers up, executes a self test, or otherwise gets ready for work.

Other features and functions include faster or slower playback—which, as usual, doesn't affect the pitch—and a switch on the side of the unit designed to bring out recorded voices after the fact. There's a minijack for headphones, and one for an external stereo microphone; the microphone jack also supplies power to microphones that require it. There's no separate line-in jack. The headphone jack is live during recording for monitoring purposes, and also raises the possibility of using the recorder as a microphone, although I haven't actually tried it myself.

As others have mentioned when reviewing this product, the case appears to be constructed of aluminum, or similar material. The only obvious use of plastic, as far as I can see, is for the battery-compartment cover on the bottom of the unit, the cover over the USB port/Memory Stick slot area, and the control buttons and switches themselves.

It comes with a slip case, which is good enough for protecting the display from scratches when it's in your pocket, and it's acoustically transparent so it can be used while recording. Ear buds are included, as well as a couple Sony AAA alkaline batteries to get you started. An AC adapter is optional/extra, but considering historical noise-related difficulties with portable recording equipment connected to AC power sources, this option may not be desirable anyway.

Taking into account the caveats mentioned—and the assumed potential for difficulty arising from large fingers on tiny controls—my overall impression of this recorder is entirely positive. Two thumbs up, and all that.


No comments:

Post a Comment