Earth Notes: Metacast 2: Podcast Specs and Workflow (2020-04)
Updated 2024-09-19.Hi, I'm Damon Hart-Davis, and welcome to Earth Notes podcast on all things eco and green and efficient @Home!
19th April 2020
This metacast is primarily driven by documenting some specs and workflow to be consistent in putting together future episodes. I'm interested in opinions on the 48kHz to 44.1kHz divide that I'm crossing too.(I promise not to do too many metacasts! Blame them on coronavirus cabin fever...)
[00:36]My dad, Adam, worked in broadcast TV. The difference between the vibrant images on studio monitors at work, and the anaemic images on TV at home, was a shock. Now I can relatively easily achieve technical audio broadcast master gamut (eg BBC Radio, at 16-bit linear PCM with 48kHz sampling) in the comfort of my own home.
The 48kHz Alliance says that
All audio production & distribution should go 48kHz. It's not about the slight superior audible quality of 48kHz. It's about being compatible with the video world ...There are advanced parts of the BBC Radio spec that I don't yet know how to achieve, such as LUFS, but most of the rest of it I can, and it will tickle me to do so!
My shiny new Blue Yeti mic supports 48kHz, ie 48ksps or kilo samples per second. I can't hear any particular difference sampling at the two speeds, and I would not expect to.
[01:43]Moving to 48ksps sampling will expand the FLAC master copy by a little under 10%. But the MP3 files are being limited by output bit rate anyway, so should not see any significant change.
There were reports a few years ago of some MP3 players not coping with anything other than 44.1ksps, but there does not seem to be anything recent. So it's likely not a significant issue now.
Thus I see no reason not to do most or all future episodes at 48ksps. And so I have set Audacity's default project rate to that as a start!
It occurs to me that if I record an episode away from home, on the MBA's built-in mic, I could consider dropping the sample rate back to 44.1ksps to save a bit of space given that the extra texture just isn't there to capture. But that seems like a faff. I'll see.
In the episode's page I have a how-to for the key elements of the workflow to prepare and publish an episode.
[02:48]At a slightly-less nitty-gritty level...
Should the podcast have standard book-ends, ie intro and exit ('outro' or coda) words and even 'jingle' or signature music?
Having canned intro and outro may make for a more consistent presentation. They may not actually save production time as something now has to be imported into the piece rather than just being said. This also risks making things feel a little less 'organic' and a little more robotic and over-professionalised.
Still, no harm in laying down a couple to try now!
[03:26]A version of the sign-off used since the podcast start (Blue Yeti mic, 48ksps):
There's more on my "Earth Notes" Web site atEarth.Org.UK.Until now each intro has has some slight variation, sometimes with a small joke embedded (coronacast, diarycast, metacast), and the date. Maybe a completely standard intro, which can be followed by a date, would be good?
[03:55]Here's the new intro used on this episode (48ksps and Blue Yeti):
Hi, I'm Damon Hart-Davis, and welcome to Earth Notes podcast on all things eco and green and efficient @Home!I expect to add new, better versions of these intros and outros, and switch to them. I may also flip between a small set of them to give episodes a little more variety, and to reduce repetition blindness [deafness?].
More material will be added to this episode's page after it goes live, such as new intros and outros, and updated workflow.
There's more on my "Earth Notes" Web site at
Earth.Org.UK.
Show Notes
The intro/outro levels don't match well the body of the episode. As a test of how things might be, I ran the MP3 through FixMyLevels Cloud-Based Loudness Correction
. FML also accepts WAV. Yes, I think the loudness is significantly improved.
It may be worth exporting as WAV, uploading to FML, then putting up FML's version of the stereo- and mono- podcast MP3s spat out.
So now my dilemma is, do I go for (dynamic) compression to get my loudness consistent and normalise in one go. I was not thinking of going full drive-time AM radio, but ...
I note that The Intelligence appears to be 128kbps CBR (44.1ksps). vs the effective 153kpbs for the primary MP3 for this episode (46kbps for the secondary MP3), which suggests that I could possibly notch mine down a bit to save bandwidth. Both are joint stereo and nominally 128kpbs (eg according to file). Though given how short the EN episodes are, no listeners are being hurt in the making of these files this way!
New Scientist Weekly is 96kbps CBR mono (44.1ksps).
How Much Speed You Need to Stream: various services' bandwidth requirements from 32kbps for Tunein spoken content, through 96--320kbps for various normal/high/best music to 5Mbps for Deezer Elite.
My Workflow: How To Make a Podcast Episode at 48ksps with Audacity
Time to create typical short ~3 minute episode, including the script and audio:
Ingredients
- Audacity 2.3.2+ (run from command line to avoid permissions problems with macOS 10.5 Catalina)
- Blue Yeti microphone
- Laptop (MacBook running macOS 10.15.4+)
- Episode script (transcript), other materials, and new episode podcast page ready to fill out
- Idle hands...
A summary of the new workflow for an audio-only episode:
-
Checklist: in Audacity, ensure that the project rate is 48000Hz, the correct microphone (usually the Blue Yeti, in cardioid mode) is selected for input, and as stereo. -
Record the main body of the episode and perform initial edits. Do some very rough edits, such as removing coughs and obvious clicks, etc. -
Apply Compress &dynamics with default settings instead (or even milder <0.5 compression), and maybe check for -1dbTP or a little under. (This may be usefully followed by Truncate Silence on default settings (-20dB threshold and truncation to 0.5s), to help move things on a bit and cut out many of the breaths, etc.) Optional (only one of this or Normalize need be done, though this followed by Normalize may be optimal). -
From late 2020 onwards... (Apply this separately to each inserted clips, intro, and outro, as well as the script, for better consistency.) Use "Loudness Normalization" (Audacity 2.4 onwards) with LUFS (perceived) of -23dB. (Plain Normalize was used up to and including this "Metacast 2" episode.) -
Edit, and import and insert other (normalised) audio segments. - Import and insert standard (already compressed/normalised) (title, sig, words) intro and (sig, words) outro blocks (from 2020-04-19 use
In-EcoGreen-48ksps-1andOut-TheresMore-48ksps-1, from 2020-05-01 usesig-5YPV-In-EcoGreen-48ksps-1andsig-5YPV-Out-TheresMore-48ksps-1). -
Re-edit the (tran)script to reflect what actually happened! -
Export master/top/archive copy as FLAC (16-bits, maximum compression), with metadata (priming with XML template from podcast-furnituredirectory if necessary). -
Export primary MP3 with Audacity nominal ~192kbps 'standard' (LAME -m j -V 2 -q 0 -lowpass 18.5 --vbr-new -b 32) preset 170--210kbps joint stereo for good voice rendering and some stereo capability. Note that this is a VBR format, so may upset some players. (Maybe this should be reduced to ~128kbps (variable 'medium' or CBR 128kbps), ie is maybe higher-fi than needed for voice.) -
Optionally export secondary MP3 with Audacity ~48kbps 'variable' (LAME -m m -V 9 -q 3 -lowpass 10 --vbr-old -b 32) 45--85kbps forced to mono, with file size one third of the primary MP3. (As of.mp3Lfiles should be auto-generated with flags-codec:a libmp3lame -qscale:a 9 -ac 1 -cutoff 10000 -compression_level 0 -f mp3and used for RSS feeds and theAUDIOtag if necessary, saving a manual step.) -
Optionally generate very-low bandwidth Opus with ffmpeg, preferably from lossless source, (ffmpeg -y -i input.flac -c:a libopus -ac 1 -b:a 16k output.opus) 16kbps mono, with file size typically one third of the secondary MP3L. (New as of ; as of one should be auto-generated with flags-codec:a libopus -ac 1 -b:a 16k -f opusand used for RSS feeds and theAUDIO/VIDEOtags if this is not done manually but it may be slightly larger and/or lower-fi on the older RPi serverffmpegversion.) -
Move the manually-created .flac,.mp3(and possibly.mp3Land.opusL) files into the appropriate episode directory underimg/audio, make sure that they are readable over HTTP, and point the episode page'sAUDIO srcat the.mp3primary. -
Transcribe (maybe with Transcobble) to a WebVTT .vttfile placed beside the primary audio/video file, eg atimg/audio/.../wit-and-wisdom.mp3.vtt. This will serve as closed captions for a video, as well as a direct podcast-friendly transcription. Also consider bringing a select few of the timing cues in as [mm:ss] (classcuetime) entries to the main page (tran)script. -
Publish the episode page and the RSS feed.
Audio Segments
Here are the segments created, to top and tail new episodes.
Intro ... all things eco ...
(48ksps, Blue Yeti mic):
Outro There's more ... at earth.org.uk
(48ksps, Blue Yeti mic).
Signature sound 5YPV — 5 years of PV generation (48ksps):
Generated .wav 60s all-zeros silence at 1Hz, resampled to 8kHz for MP3s: minimum 8kbps (1kBps/min) for mono .mp3L, 8kbps (1.5kBps/min) for joint-stereo .mp3, and 5bps (!) for .flac at usual settings:
"Diarycast" (48ksps, Blue Yeti mic):
"Statscast" (48ksps, Blue Yeti mic):
"Metacast" (48ksps, Blue Yeti mic):
Silly
And some silly noises, for the right moments...
Raspberry:
Footsteps:
Possible Episode Structure
- Spoken headline (possibly simplified from that of the page, or a fixed 'type' such as "diarycast", "statscast" or "metacast")
- Signature tune/sound (~2s/5s/10s), with possible fade into...
- Standard spoken intro
- Date
- BODY, maybe starting with page description and key para, for content pyramid
- Signature tune/sound (~5s), with possible fade into...
- Standard spoken outro
2020-04-25: Zencastr Test
I'm testing viable ways to do a remote podcast interview. One such is Zencastr. At the moment (during corona troubles) the freebie 'Hobbyist' tier has no usage limits and produces a decent mono 44.1ksps 128kbps CBR MP3 track for each participant. These are downloaded separately at the end of the session. I can then load these in Audacity beside one another and mix down to a new combined track, possibly after tweaking each first.
Gethin D very kindly agreed to a test session with me and here are the raw MP3s captured locally one each side. Before and after the 'interview' we could hear each other through Zencastr's (good) VoIP, making it easy for me to decide when to hit the 'record' button to start the session proper.
Here is each (mono) track, me on Blue Yeti and Gethin on his headset mic.
And here is them with some noise silencing at the start, "Compression Dynamics" applied, and then mixed together and rendered into a new stereo track. I'm panned 20% left and Gethin 20% right (maybe). There is no FLAC for this, but I have created the forced-mono low-fi version.
(Panning Instruments has some suggests about panning in music.)
When making a full episode containing an interview I could upsample these into a 48ksps project or run the whole piece at 41.1ksps, thus downsampling the intro/outro instead. If I go 'professional' with Zencastr I get WAVs which would remove one minor level of conversion...
(Maybe I should stay centre but just pan guests slightly to left and right to make good use of joint stereo for at least my part, and for more consistency with intro/outro.)
2020-05-25: Zoom H1n/UK Portable Mic
Just in case more ambient or non-desk-based recording might be on the cards, and given how unportable the Blue Yeti is for that purpose, I have considered the Zoom H1n/UK Handy Recorder (£80 including VAT).
- It was suggested by friends, and Zoom seems to have a good reputation.
- It covers the stereo, 16-bit, 48ksps format that I's like to use.
- It captures in WAV and MP3.
- I'd need to buy a 32GB (max) microSDHC card (or repurpose one).
- It takes 2 x AAA cells. I was able to verify that it will take NiMH rechargeables.
- My understanding is that a microUSB can be used to off-load tracks.
I would get a matching windscreen / pop-filter if I bought this.
I don't expect to buy one any time soon, but when I get to travel and socialise again post lockdown, I'll consider it!
2020-12-22: I caved and bought one!
2020-05-27: AudioMoth
Another possibility, which would allow set-and-forget collection of sound at up to bat ultrasonic frequencies, is the AudioMoth by OAD.
2021-01-19: yes, I bought one of these too!
2020-06-08: LUFS
I have just noticed that Audacity 2.4.1 under Effect (Built-in) has a "Loudness Normalization" that defaults to -23dB (per BBC standards).
2020-08-17: new episodes are having that applied as a final step, though the result can feel a bit too quiet...