May 262020
 

Lately, I’ve been socially distancing a home and so there’s been a few projects that have been considered that otherwise wouldn’t ordinarily get a look in on a count of lack-of-time.

One of these has been setting up a Raspberry Pi with DRAWS board for use on the bicycle as a radio interface. The DRAWS interface is basically a sound card, RTC, GPS and UART interface for radio interfacing applications. It is built around the TI TMS320AIC3204.

Right now, I’m still waiting for the case to put it in, even though the PCB itself arrived months ago. Consequently it has not seen action on the bike yet. It has gotten some use though at home, primarily as an OpenThread border router for 3 WideSky hubs.

My original idea was to interface it to Mumble, a VoIP server for in-game chat. The idea being that, on events like the Yarraman to Wulkuraka bike ride, I’d fire up the phone, connect it to an AP run by the Raspberry Pi on the bike, and plug my headset into the phone:144/430MHz→2.4GHz cross-band.

That’s still on the cards, but another use case came up: digital. It’d be real nice to interface this over WiFi to a stronger machine for digital modes. Sound card over network sharing. For this, Mumble would not do, I need a lossless audio transport.

Audio streaming options

For audio streaming, I know of 3 options:

  • PulseAudio network streaming
  • netjack
  • trx

PulseAudio I’ve found can be hit-and-miss on the Raspberry Pi, and IMO, is asking for trouble with digital modes. PulseAudio works fine for audio (speech, music, etc). It will make assumptions though about the nature of that audio. The problem is we’re not dealing with “audio” as such, we’re dealing with modem tones. Human ears cannot detect phase easily, data modems can and regularly do. So PA is likely to do things like re-sample the audio to synchronise the two stations, possibly use lossy codecs like OPUS or CELT, and make other changes which will mess with the signal in unpredictable ways.

netjack is another possibility, but like PulseAudio, is geared towards low-latency audio streaming. From what I’ve read, later versions use OPUS, which is a no-no for digital modes. Within a workstation, JACK sounds like a close fit, because although it is geared to audio, its use in professional audio means it’s less likely to make decisions that would incur loss, but it is a finicky beast to get working at times, so it’s a question mark there.

trx was a third option. It uses RTP to stream audio over a network, and just aims to do just that one thing. Digging into the code, present versions use OPUS, older versions use CELT. The use of RTP seemed promising though, it actually uses oRTP from the Linphone project, and last weekend I had a fiddle to see if I could swap out OPUS for linear PCM. oRTP is not that well documented, and I came away frustrated, wondering why the receiver was ignoring the messages being sent by the sender.

It’s worth noting that trx probably isn’t a good example of a streaming application using oRTP. It advertises the stream as G711u, but then sends OPUS data. What it should be doing is sending it as a dynamic content type (e.g. 96), and if this were a SIP session, there’d be a RTPMAP sent via Session Description Protocol to say content type 96 was OPUS.

I looked around for other RTP libraries to see if there was something “simpler” or better documented. I drew a blank. I then had a look at the RTP/RTCP specs themselves published by the IETF. I came to the conclusion that RTP was trying to solve a much more complicated use case than mine. My audio stream won’t traverse anything more sophisticated than a WiFi AP or an Ethernet switch. There’s potential for packet loss due to interference or weak signal propagation between WiFi nodes, but latency is likely to remain pretty consistent and out-of-order handling should be almost a non-issue.

Another gripe I had with RTP is its almost non-consideration of linear PCM. PCMA and PCMU exist, 16-bit linear PCM at 44.1kHz sampling exists (woohoo, CD quality), but how about 48kHz? Nope. You have to use SDP for that.

Custom protocol ideas

With this in mind, my own custom protocol looks like the simplest path forward. Some simple systems that used by GQRX just encapsulate raw audio in UDP messages, fire them at some destination and hope for the best. Some people use TCP, with reasonable results.

My concern with TCP is that if packets get dropped, it’ll try re-sending them, increasing latency and never quite catching up. Using UDP side-steps this, if a packet is lost, it is forgotten about, so things will break up, then recover. Probably a better strategy for what I’m after.

I also want some flexibility in audio streams, it’d be nice to be able to switch sample rates, bit depths, channels, etc. RTP gets close with its L16/44100/2 format (the Philips Red-book standard audio format). In some cases, 16kHz would be fine, or even 8kHz 16-bit linear PCM. 44.1k works, but is wasteful. So a header is needed on packets to at least describe what format is being sent. Since we’re adding a header, we might as well set aside a few bytes for a timestamp like RTP so we can maintain synchronisation.

So with that, we wind up with these fields:

  • Timestamp
  • Sample rate
  • Number of channels
  • Sample format

Timestamp

The timestamp field in RTP is basically measured in ticks of some clock of known frequency, e.g. for PCMU it is a 8kHz clock. It starts with some value, then increments up monotonically. Simple enough concept. If we make this frequency the sample rate of the audio stream, I think that will be good enough.

At 48kHz 16-bit stereo; data will be streaming at 192kbps. We can tolerate wrap-around, and at this data rate, we’d see a 16-bit counter overflow every ~341ms, which whilst not unworkable, is getting tight. Better to use a 32-bit counter for this, which would extend that overflow to over 6 hours.

Sample rate encoding

We can either support an integer field, or we can encode the rate somehow. An integer field would need a range up to 768k to support every rate ALSA supports. That’s another 32-bit integer. Or, we can be a bit clever: nearly every sample rate in common use is a harmonic of 8kHz or 11.025kHz, so we devise a scheme consisting of a “base” rate and multiplier. 48kHz? That’s 8kHz×6. 44.1kHz? That’s 11.025kHz×4.

If we restrict ourselves to those two base rates, we can support standard rates from 8kHz through to 1.4MHz by allocating a single bit to select 8kHz/11.025kHz and 7 bits for the multiplier: the selected sample rate is the base rate multiplied by the multipler incremented by one. We’re unlikely to use every single 8kHz step though. Wikipedia lists some common rates and as we go up, the steps get bigger, so let’s borrow 3 multiplier bits for a left-shift amount.

7 6 5 4 3 2 1 0
B S S S M M M M

B = Base rate: (0) 8000 Hz, (1) 11025 Hz
S = Shift amount
M = Multiplier - 1

Rate = (Base << S) * (M + 1)

Examples:
  00000000b (0x00): 8kHz
  00010000b (0x10): 16kHz
  10100000b (0xa0): 44.1kHz
  00100000b (0x20): 48kHz
  01010010b (0x52): 768kHz (ALSA limit)
  11111111b (0xff): 22.5792MHz (yes, insane)

Other settings

I primarily want to consider linear PCM types. Technically that includes unsigned PCM, but since that’s losslessly transcodable to signed PCM, we could ignore it. So we could just encode the number of bytes needed for a single channel sample, minus one. Thus 0 would be 8-bits; 1 would be 16-bits; 2 would be 32-bits and 3 would be 64-bits. That needs just two bits. For future-proofing, I’d probably earmark two extra bits; reserved for now, but might be used to indicate “compressed” (and possibly lossy) formats.

The remaining 4 bits could specify a number of channels, again minus 1 (mono would be 0, stereo 1, etc up to 16).

Packet type

For the sake of alignment, I might include a 16-bit identifier field so the packet can be recognised as being this custom audio format, and to allow multiplexing of in-band control messages, but I think the concept is there.

May 222020
 

For the past 2 years now, there’s been quite a bit in the press about the next evolution of mobile telephony standards.

The 5G standard is supposed to bring with it higher speeds and greater user density handling. As with a lot of systems, “5G” itself, describes a family of standards… some concern the use of millimetre-wave communications for tower-to-handset communications, some cover the communications channels for more modest frequencies in the high UHF bands.

One thing that I really can’t get my head around is the so-called claims of health effects.

Now, these are as old as radio communications itself. And for sure, danger to radio transmissions does increase with frequency, proximity and transmit power. There is a reason why radio transmitter sites such as those that broadcast medium wave radio or television are fenced off: electrocution is a real risk at high power.

0G: glorified two-way radios

Mobile phones originally were little more than up-market cordless phones. They often were a luggable device if they were portable at all. Many were not, they were installed into a vehicle (hence “mobile”). No such thing as cell hand-over, and often incoming calls had to be manually switched.

Often the sets were half-duplex, and despite using a hand-set, would have a very distinctive “radio” feel to them, requiring the user use a call-sign when initiating a call, and pressing a push-to-talk button to switch between listening and talking modes.

These did not see much deployment outside the US or maybe Europe.

1G: cellular communications

Back in the late 80s, when AMPS mobile phones (1G) were little more than executive toys, there might not have been much press about, but I’m sure there’d be anecdotal evidence of people being concerned about “radiation”.

If any standard was going to cause problems, it’d have been 1G, since the sets generally used much higher transmit power to compensate for the lack of coverage. They were little more than glorified FM transceivers with a little digital control channel on the side which implemented the selective calling and cell hand-off.

This was the first standard we saw here in Australia, and was the first to be actually practical. Analogue services didn’t last that long, and because of the expense of running AMPS services, they were mostly an expensive luxury. So that did limit its up-take.

2G: voice goes digital

The next big change was 2G, which replaced the analogue FM voice channel and used digital modulation techniques. GSM (which used Gaussian Minimum Shift Keying) and CDMA (which used phase shift keying) encoded everything in a single digital transmission.

This meant audio could be compressed (with some loss in fidelity), and have forward error correction added to make the signal more robust to noise. The cells could handle more users than the 1G services could. Transmit power could be reduced, improving battery life and the sets became cheaper to make and services became more economical.

Then came all the claims that 2G was going to cause us to develop brain cancer.

Now, many of those 2G services started popping up in the mid 90s… has there been a mass pandemic of cancer cases? Nope! About the only thing GSM was bad for, was its ability to leak into any audio frequency circuit.

2G went through a few sub-revisions, but it basically was AMPS done digitally, so fundamentally worked much the same. A sore point was how data was handled. 2G and its predecessors all tried to emulate what the wired network was doing: establishing a dedicated circuit between callers.

The Internet was really starting to get popular, and people wanted a way to access it on the move. GPRS did allow for some of that, but it really didn’t work that well due to the way 2G saw the world, so things moved on.

3G: packet switching

The big change here was the move from “circuits” to sending data around in packets. This is more like how the Internet operates, and so it meant the services could better support an Internet connection.

Voice still went the old-fashioned way, dedicated circuits, since the QoS (quality of service) could be better maintained that way.

The cells could support more users than 2G could, and the packet mode meant mobile Internet finally became a “thing” for most people.

I don’t recall there being the same concern about health as there was for 2G… it was probably still simmering below the surface. Services were deployed further afield and of course, the uptake continued.

4G: bye bye circuit switching

4G or LTE is the current standard that most of us are using. The biggest change is it ditches the circuit switching used in 1G, 2G and 3G. Voice is done using VoLTE… basically the voice call is sent the same way calls are routed over the Internet.

The cell towers are no longer trying to keep a “circuit” connected to your phone as you move around, instead it’s just directing packets. It’s your handset’s problem to sort out whether it heard a given packet already, or re-arrange incoming packets if they arrive out-of-order.

To make this work, obviously the latency inherent in 3G had to be addressed. As a sweetener, the speeds were bumped up, and the voice CODEC could be updated, so we gained wide-band voice calls. (Pity Bluetooth hasn’t kept up!)

5G: new frequencies, higher speed, smaller cells

So far, the cellular standards have largely co-existed in the same frequency bands. 4G actually varies quite a bit in frequency, but basically there are bands from the low UHF around 410MHz right up to microwave at 2600MHz.

Higher frequencies

5G has been contentious because some implementations of it reach even higher. Frequency Range 1 used in the 5G NR standard is basically much the same as 4G, but frequency range 2 soars as high as 40GHz.

Now, in terms of the electromagnetic spectrum, compared to other forms of radiation that we rely on for survival (and have done ever since life first began on this planet), this might as well be DC!

Infrared radiation, which is the very bottom of the “light” spectrum, starts at 300GHz. At these frequencies, we typically forget about frequencies, and instead consider wavelengths (1mm in this case). Visible light is even higher, 430THz (yes, that’s T for tera!).

Now, where do we start to worry about radiation? The nasty stuff begins with ultraviolet radiation, specifically UVC which is at a dizzying 1.1PHz (yes, that’s peta-hertz). It’s worth noting that UVB, which is a little lower in frequency can cause problems when exposure is excessive… however none is dangerous too, you actually need UVB exposure on your body to produce vitamin D for survival!

Dielectric heating

So that’s where the danger is in terms of frequency. I did mention that danger also increases with power… this is why microwave ovens, which typically operate at a fairly modest 2.4GHz frequency, pose a risk.

No, they won’t make you develop cancer, but the danger there is when there’s a lot of power, it can cause dielectric heating. That is, it causes molecules to move around, and in doing so, collide transferring energy which is then given off as heat. It happens at all frequencies in the EM spectrum, but it starts to become more practical at microwave frequencies.

To do something like cook dinner, a microwave oven bombards your food with hundreds of watts of RF energy at it. The microwave has a thick RF shield around it for a reason! If that shield is doing what it should, you might be exposed to no more than a watt of energy escaping the shield. Not enough to cause any significant heating.

I hear that if you put a 4W power amp on a 2.4GHz WiFi access point and put your hand in front of the antenna, you can “feel” framing packets. (Never tried this myself.) That’s pretty high power for most microwave links, and would be many orders of magnitude more than what any cell phone would be capable of.

Verdict: not a health risk

In my view, there’s practically no risk in terms of health effects from 5G. I expect my reasoning above will be thoroughly rubbished by those who are protesting against the roll-out.

However, that does not mean I am in favour of 5G.

The case against 5G

So I seem to be sticking up for 5G above, but let me make one thing abundantly clear, for us here in Australia, I do not think 5G is the “right” thing for us to use. It’s perfectly safe in terms of health effects, but simply the wrong tool for the job.

Small cells

Did I mention before the cells were smaller? Compared to its predecessors, 5G cells are tiny! The whole point of 5G was to serve a large number of users in a small area. Think of 10s of thousands of people crammed into a single stadium (okay, once COVID-19 is put to bed). That’s the use case for 5G.

5G’s range when deployed on the lower bands, is about on par with 4G. Maybe a little better in certain ideal conditions with higher speeds. This is likely the variant we’re most likely to see outside of major city CBDs. How reliable it is at that higher speed remains to be seen, as there’s a crazy amount of DSP going on to make stuff work at those data rates.

5G when deployed with mmWave bands, barely makes 500 metres. This will make deployment in the suburbs prohibitively expensive. Outdoor Wi-Fi or WiMAX might not be as fast, but would be more cost-effective!

Processor load

Did I mention about the crazy amount of DSP going on? To process data streams that exceed 1Gbps, you’re doing a lot of processing to extract the data out of the radio signal. 5G leans heavily on MIMO for its higher speeds, basically dividing the high-rate stream into parts which are directed to separate antennas. This reduces the bandwidth needed to achieve a high data rate, but it does make processing the signal at the far end more complex.

Consequently, the current crop of 5G handsets run hot. How hot? Well, subject them to 29.5°C, and they shut down! Now, think about the weather we get in this country? How many days have we experienced lately where 29°C has been a daily minimum, not a maximum?

5G isn’t the future for Australia

We need a wireless standard that goes the distance, and can take the heat! 5G is not looking so great in this marathon race. Personally, I’d like to see more investment into the 4G services and getting those rolled out to more locations. There’s plenty of locations that are less than a day’s drive from most capital cities, where mobile coverage is next to useless.

Plenty of modern 4GX handsets also suffer technical elitism… they see 3G services, but then refuse to talk to them, instead dropping to -1G: brick emulation. There’s a reason I stick by my rather ancient ZTE T83 and why I had high hopes for the Kite.

I think for the most part, many of the wireless standards we see have been driven by Europe and Asia, both areas with high population densities and relatively cool annual temperatures.

It saddens me when I hear Telstra tell everybody that they “aspire” to be a technology company, when back in the early 90s, Telecom Australia very much was a technology company, and a well respected trail-blazing one at that! It’s time they pulled their finger out and returned to those days.

May 122020
 

So, the other day I pondered about whether BlueTrace could be ported to an older device, or somehow re-implemented so it would be compatible with older phones.

The Australian Government has released their version of TraceTogether, COVIDSafe, which is available for newer devices on the Google and Apple application repositories. It suffers a number of technical issues, one glaring one being that even on devices it theoretically supports, it doesn’t work properly unless you have it running in the foreground and your phone unlocked!

Well, there’s a fail right there! Lots of people, actually need to be able to lock their phones. (e.g. a condition of their employment, preventing pocket dials, saving battery life, etc…)

My phone, will never run COVIDSafe, as provided. Even compiling it for Android 4.1 won’t be enough, it uses Bluetooth Low Energy, which is a Bluetooth 4.0 feature. However, the government did one thing right, they have published the source code. A quick fish-eye over the diff against TraceTogether, suggests the changes are largely superficial.

Interestingly, although the original code is GPLv3, our government has decided to supply their own license. I’m not sure how legal that is. Others have questioned this too.

So, maybe I can run it after all? All I need is a device that can do BLE. That then “phones home” somehow, to retrieve tokens or upload data. Newer phones (almost anything Android-based) usually can do WiFi hotspot, which would work fine with a ESP32.

Older phones don’t have WiFi at all, but many can still provide an Internet connection over a Bluetooth link, likely via the LAN Access Profile. I think this would mean my “token” would need to negotiate HTTPS itself. Not fun on a MCU, but I suspect someone has possibly done it already on ESP32.

Nordic platforms are another option if we go the pure Bluetooth route. I have two nRF52840-DK boards kicking around here, bought for OpenThread development, but not yet in use. A nicety is these do have a holder for a CR2032 cell, so can operate battery-powered.

Either way, I think it important that the chosen platform be:

  1. easily available through usual channels
  2. cheap
  3. hackable, so the devices can be re-purposed after this COVID-19 nonsense blows over

A first step might be to see if COVIDSafe can be cleaved in two… with the BLE part running on a ESP32 or nRF52840, and the HTTPS part running on my Android phone. Also useful, would be some sort of staging server so I can test my code without exposing things. Not sure if there is such a beast publicly available that we can all make use of.

Guess that’ll be the next bit to look at.

May 042020
 

Sure, one moment, let’s try your link…

Errm “No such app found”… I think your link is broken guys, please fix! Bear in mind, my phone is one of these. It still makes calls, still sends and receives text messages, still does what I need it to do.

If it doesn’t do what you need it to do, that is not my problem, take that up with Telstra/ZTE.

May 032020
 

This afternoon, I was pondering about how I might do text-to-speech, but still have the result sound somewhat natural. For what use case? Well, two that come to mind…

The first being for doing “strapper call” announcements at horse endurance rides. A horse endurance ride is where competitors and their horses traverse a long (sometimes as long as 320km) trail through a wilderness area. Usually these rides (particularly the long ones) are broken up into separate stages or “legs”.

Upon arrival back at base, the competitor has a limited amount of time to get the horse’s vital signs into acceptable ranges before they must present to the vet. If the horse has a too-high temperature, or their horse’s heart rate is too high, they are “vetted out”.

When the competitor reaches the final check-point, ideally you want to let that competitor’s support team know they’re on their way back to base so they can be there to meet the competitor and begin their work with the horse.

Historically, this was done over a PA system, however this isn’t always possible for the people at base to achieve. So having an automated mechanism to do this would be great. In recent times, Brisbane WICEN has been developing a public display that people can see real-time results on, and this also doubles as a strapper-call display.

Getting the information to that display is something of a work-in-progress, but it’s recognised that if you miss the message popping up on the display, there’s no repeat. A better solution would be to “read out” the message. Then you don’t have to be watching the screen, you can go about your business. This could be done over a PA system, or at one location there’s an extensive WiFi network there, so streaming via Icecast is possible.

But how do you get the text into speech?

Enter flite

flite is a minimalist speech synthesizer from the Festival project. Out of the box it includes 3 voices, mostly male American voices. (I think the rms one might be Richard M. Stallman, but I could be wrong on that!) There’s a couple of demos there that can be run direct from the command line.

So, for the sake of argument, let’s try something simple, I’ll use the slt voice (a US female voice) and just get the program to read out what might otherwise be read out during a horse ride event:

$ flite_cmu_us_slt -t 'strapper call for the 160 kilometer event competitor numbers 123 and 234' slt-strapper-nopunctuation-digits.wav
slt-strapper-nopunctuation-digits.ogg

Not bad, but not that great either. Specifically, the speech is probably a little quick. The question is, how do you control this? Turns out there’s a bit of hidden functionality.

There is an option marked -ssml which tells flite to interpret the text as SSML. However, if you try it, you may find it does little to improve matters, I don’t think flite actually implements much of it.

Things are improved if we spell everything out. So if you instead replace the digits with words, you do get a better result:

$ flite_cmu_us_slt -t 'strapper call for the one hundred and sixty kilometer event competitor number one two three and two three four' slt-strapper-nopunctuation-words.wav
slt-strapper-nopunctuation-words.ogg

Definitely better. It could use some pauses. Now, we don’t have very fine-grained control over those pauses, but we can introduce some punctuation to have some control nonetheless.

$ flite_cmu_us_slt -t 'strapper call.  for the one hundred and sixty kilometer event.  competitor number one two three and two three four' slt-strapper-punctuation.wav
slt-strapper-punctuation.ogg

Much better. Of course it still sounds somewhat robotic though. I’m not sure how to adjust the cadence on the whole, but presumably we can just feed the text in piece-wise, render those to individual .wav files, then stitch them together with the pauses we want.

How about other changes though? If you look at flite --help, there is feature options which can control the synthesis. There’s no real documentation on what these do, what I’ve found so far was found by grep-ing through the flite source code. Tip: do a grep for feat_set_, and you’ll see a whole heap.

Controlling pitch

There’s two parameters for the pitch… int_f0_target_mean controls the “centre” frequency of the speech in Hertz, and int_f0_target_stddev controls the deviation. For the slt voice, …mean seems to sit around 160Hz and the deviation is about 20Hz.

So we can say, set the frequency to 90Hz and get a lower tone:

$ flite_cmu_us_slt --setf int_f0_target_mean=90 -t 'strapper call' slt-strapper-mean-90.wav
slt-strapper-mean-90.ogg

… or 200Hz for a higher one:

$ flite_cmu_us_slt --setf int_f0_target_mean=200 -t 'strapper call' slt-strapper-mean-200.wav
slt-strapper-mean-200.ogg

… or we can change the variance:

$ flite_cmu_us_slt --setf int_f0_target_stddev=0.0 -t 'strapper call' slt-strapper-stddev-0.wav
$ flite_cmu_us_slt --setf int_f0_target_stddev=70.0 -t 'strapper call' slt-strapper-stddev-70.wav
slt-strapper-stddev-0.ogg
slt-strapper-stddev-70.ogg

We can’t change these values during a block of speech, but presumably we can cut up the text we want to render, render each piece at the frequency/variance we want, then stitch those together.

Controlling rate

So I mentioned we can control the rate, somewhat coarsely using usual punctuation devices. We can also change the rate overall by setting duration_stretch. This basically is a control of how “long” we want to stretch out the pronunciation of words.

$ flite_cmu_us_slt --setf duration_stretch=0.5 -t 'strapper call' slt-strapper-stretch-05.wav
$ flite_cmu_us_slt --setf duration_stretch=0.7 -t 'strapper call' slt-strapper-stretch-07.wav
$ flite_cmu_us_slt --setf duration_stretch=1.0 -t 'strapper call' slt-strapper-stretch-10.wav
$ flite_cmu_us_slt --setf duration_stretch=1.3 -t 'strapper call' slt-strapper-stretch-13.wav
$ flite_cmu_us_slt --setf duration_stretch=2.0 -t 'strapper call' slt-strapper-stretch-20.wav
slt-strapper-stretch-05.ogg
slt-strapper-stretch-07.ogg
slt-strapper-stretch-10.ogg
slt-strapper-stretch-13.ogg
slt-strapper-stretch-20.ogg

Putting it together

So it looks as if all the pieces are there, we just need to stitch them together.

RC=0 stuartl@rikishi /tmp $ flite_cmu_us_slt --setf duration_stretch=1.2 --setf int_f0_target_stddev=50.0 --setf int_f0_target_mean=180.0 -t 'strapper call' slt-strapper-call.wav
RC=0 stuartl@rikishi /tmp $ flite_cmu_us_slt --setf duration_stretch=1.1 --setf int_f0_target_stddev=30.0 --setf int_f0_target_mean=180.0 -t 'for the, one hundred, and sixty kilometer event' slt-160km-event.wav
RC=0 stuartl@rikishi /tmp $ flite_cmu_us_slt --setf duration_stretch=1.4 --setf int_f0_target_stddev=40.0 --setf int_f0_target_mean=180.0 -t 'competitors, one two three, and, two three four' slt-competitors.wav
Above files stitched together in Audacity

Here, I manually imported all three files into Audacity, arranged them, then exported the result, but there’s no reason why the same could not be achieved by a program, I’m just inserting pauses after all.

There are tools for manipulating RIFF waveform files in most languages, and generating silence is not rocket science. The voice itself could be fine-tuned, but that’s simply a matter of tweaking settings. Generating the text is basically a look-up table feeding into snprintf (or its equivalent in your programming language of choice).

It’d be nice to implement a wrapper around flite that took the full SSML or JSML text and rendered it out as speech, but this gets pretty close without writing much code at all. Definitely worth continuing with.