The Secret Livesof MP3 Files

Doug KayeThe Conversations Network

and GigaVox Media

Formats & Encoders

• Lossless (WAV, AIFF)

• Lossy

- MPEG 1, Layer 3 (MP3)

- AAC (AAC, M4A, M4B)

- MPEG I, Layer 2 (MP2)

MPEG Confusion

• Lossy Perceptual/Psychoacoustical Codecs

• MP3 = MPEG-I Layer 3

• MP2 = MPEG-I Layer 2 (not MPEG-II)

Motion Picture Experts Group

• MPEG-1: Video CDs, MP3 Audio

• MPEG-2: Digital TV, Set-Top Boxes

• MPEG-4: Online Multimedia (Video)

• MPEG-7: Audio and Video Search

• MPEG-21: Multimedia Framework

MPEG-1 for Geeks

• Layer 1

• Simple 32-Band Algorithm

• Philips DCC (Digital Compact Cassette)

• Layer 2 (a.k.a. MUSICAM)

• Also 32 Bands

• International Standard for Broadcasting

MPEG-1 Layer 3 (MP3)for Geeks

• Psychoacoustic Masking

• 32 Bands Divided into 576 Subbands

• More Accurate Masking Thresholds

• Redundancy Reduction

• Lossless Huffman Encoding

• Bit-Reservoir Buffering

• Joint Stereo

Sample Rate for Geeks

• The Nyquist Theorem

• Sample at 2x the Highest Frequency

• 22.05kHz Sample Rate for 11kHz Audio

• Sample Rate Is a Property of Uncompressed Source (WAV or AIFF)

Sample Rate in Practice

• Standardize on 44.1kHz Sample Rate

• Flash & Other Players Require n*11.025kHz

• Resample if Source is 48kHz from DVDs

Bit Rate for Geeks

• Independent of Sample Rate

• Specifies Encoder Output File Size (CBR)

• @64kbps, 1 hour ≈ 27MB

• Variable Bit Rate (VBR)

• For Higher Bit Rates Only

• Not Universally Supported (Avoid It)

Bit Rate in Practice

• “Use Higher Bit Rates for Music?”

• It’s a Myth!

• Human Voices Are Complex

• Music Masks Its Own Artifacts

• 64kbps is Most Common Today

• 96kbps is Gaining

Podcasting Bit-Rate History

• June 2003: 32kbps. “Files too large”

• April 2004: 48kbps. “No problem”

• September 2004: 64kbps. “Quality is low”

• Today: Still 64kbps.

• Tomorrow??

Stereo Encoding

• “Stereo MP3s are twice as large as mono.”

• It’s a Myth!

• Only Bit Rate Specifies Output File Size

• You May Want to Use Higher Bit Rates for Stereo

Stereo Encoding for Geeks

• Dual Channel or Independent Channel (IC)

- Entirely Separate Left and Right

• But Most L/R Information is Redundant

• Intensity Stereo (IS)

• Mid/Side Stereo (MS)

• Joint Stereo (JS) Allows IS/MS Combination

Stereo Encoding(Even Geekier)

• JS Encodes L+R and L-R

• If L=R then L-R=0

• Since Bit Rate is ConstantL=R Uses Fewer Bits for Stereo Information

Stereo Encoding in Practice

• Stereo vs. Mono (not Music vs. Voice) is a Good Reason to Use Higher Bit Rates

• Greater Separation Suggests Higher Rates

• If Mostly Speech, Consider 100% Mono

• If Mono, Make L&R Digitally Identical

• Always Encode in Stereo for Compatibility

Mastering for MP3

• Help the Encoder: Eliminate Unnecessary Data

- High-Pass Filter at 80Hz

- Low-Pass Filter at 11kHz (@64kbps encoding)

- Normalize

Which is Louder?

• It’s Not the Height of the Peaks (voltage)

• It’s the Area Under the Curve (power)

Loudness

• What’s the Standard?

• We Asked:

- Podcasters

- Audio Engineers

- Radio Engineers

• Answer: There Isn’t One

• It’s a Hard Problem to Solve

Normalization

• Peak Normalization (common)

- Maximizes Voltage, not Power

• RMS Normalization

- Maximizes Power (=Loudness)

• Determine a Standard Loudness Level

Avoid Recording to MP3!

• MP3 is a final/release format.

• Not designed to be decoded and re-encoded.

• Use MP2 Instead...

• or the highest MP3 bit rate possible.

AAC/M4B Files?

• Yes, AAC is Better Than MP3

• We Added AAC to Support iPod Bookmarks

• Painful: Only iTunes Could Encode M4B

• Doubled Much of Our Workflow

• Can’t Be Easily Assembled

MP2: Why and When?

• MPEG-1 Layer 2

• Designed as an Intermediate Format

• The Standard in Broadcast Radio

• 128kbps per Track

• 44.1kHz Sample Rate Preferred

Audio Lessons Learned

• MP3 Options

• Audio-File Myths

• RMS Normalization (Loudness)

• AAC/M4B Files (iTunes & iPods)

• MP2 Files

To Summarize

• Record at 44.1kHz Sample Rate (not in MP3!)

• Mastering

- RMS Normalization (Pick a Standard Level)

- 80Hz Hi-Pass, 11kHz Low Pass (for voice)

- If Mono, Make L&R Digitally Identical

• Encoding

- 64kbps when L=R

- Consider ≥96kbps for L≠R

- Always Use Joint Stereo