QoS

 
 

03-04-2016 15:55:17

19-12-2018 - 11:3621-04-2019 - 11:19

About TRQL Radio

An Agile Story/Timeline

2015 : the idea to create a radio comes to the mind of Pat Boens. No target; just a wish! No time to devote to this project yet. Only a nice wish that is simply postponed.

February 2016 : the idea to create a radio turns into a profound wish. No target; no requirement except that it will be a radio whose unique show will take place on a Friday, a Saturday or Sunday. Not more than 4 hours; most probably between 2 and 4 hours.

May 2016 : it will be a webradio. Its musical format will be Adult Contemporary Music, all day long!

June 2016 : music will rotate on a revolving clock of 5 days; first stats: an Adult Contemporary Music track is about 4'2" on average. 24 hours of music with no repetition requires 357 tracks. 5 days of music without repetition requires 1785 tracks. Pat Boens' digital music library is about 5000 tracks. The radio seems possible. The revolving clock will evolve with time from 5 days to 15 days.

September 2016 : the digital music library of Pat Boens is more about 15000 tracks than 5000. The wish grows into a personal necessity, a deep call, a sense of purpose.

October 29th, 2016 : Pat Boens announces to friends his intention to create his own radio. It will be called TRQL, which stands for Tranquille (Tranquil). It will be an Adult Contemporary Radio Station. I will be put LIVE on March 17, 2017.

January, 2017 : the first official logo is created by Cédric Le Dantec. A Track Wishlist is created and quickly reaches 1500 titles. The decision is taken to write ALL the computer code needed for the radio (based on the study of existing solutions).

March 17th, 2017 : the launch does not happend! Too much work to do in the field of Digital Transformations in software arenas: consulting job of Pat Boens.

April, 2017 : holidays in Thailand, Koh Samui: the first algorithm descriptions written in a notebook. This is the real start of the adventure.

April 17th, 2017 : the first TRQL Radio Player is created. It is capable to chain 8 tracks, after which it crashes.

April 21th, 2017 : the first fully functional TRQL Radio Player is published on Facebook.

It took us 14 days to go from nothing to a first player (very first version of the main program for TRQL Radio). How agile that is?
Publication on Facebook: 17:34

June, 2017 : the Player is robust enough to chain dozens of tracks before crashing: it is mainly a question of catalog – data; not a question of code.

July 5th, 2017 : the TRQL Group is created on Facebook. We have reached 5 days of music. It took us 89 days to go from nothing to something we could air, offering 5 days of music, developing EVERYTHING ourselves. You said Agile? Of course not all the Business Value could be delivered from day 1, but at least we had something real, something we could hear, making sure we could take new decisions based on facts and not on plans.

November 21st, 2017 : 19 days of music (± 6800 tracks).

December 1st, 2017 : Official start of the radio for Sabam.

December, 2017 : First effective algorithms with plain Artificial Intelligence to create the playlists (after a lot of trials and errors that required manual corrections).

December 24th and 25th, 2017 : First XMAS' days ever for TRQL Radio: a very special mix, all made manually. On 25th: first tribute ever for … George Michael: all made manually.

December 31st, 2017 : First New Year's Eve for TRQL Radio: a special mix, all made manually.

January 10st, 2018 : second tribute (all manually) for David Bowie. First information harvest for birth and death (so that tributes can be automated, at least partially). XMas tracks still appear in programs.

February, 2018 : First lyrics collected about the tracks owned in the playlists of the radio. First AI algorithms to "understand" the lyrics (so that tracks can be put in relation based on the themes they tackle). First Valentine mix of TRQL Radio: all composed manually.

March, 2018 : from 250 members to 445 in our Facebook group. Loss of 2300 tracks in the playlists. Lyrics understanding on-hold. First developments for the streaming of TRQL Radio. Later in March : Playlists are turned into a music catalog (9283 tracks officially). First playlists generated by AI (use of the global catalog). Streaming not working yet.

March 24th, 2018 : Streaming in test. This goes hand in hand with our 1st sponsor. Facing the need to display jingles of sponsoring on a regular basis. This impacts the creation of automatic playlists. 9837 tracks in our catalog. Current AI algorithm: #20.

April, 2018 : setting up a Listening Committee. 695 members in the FB group. Lot of new jingles. Streaming still not fully OK. TRQL Radio is the music playing in the background by the "Seniors antenna" of the city of Brussels (official opening presented by Alain Courtois and Mohamed Ouriaghli - Louise and La Rotonde).

April 25th, 2018 : Streaming OK for 12 hours, then crashes. From algorithm #20 to algorithm #24 (with all intermediate steps). 693 members on Facebook: target set to 1000 members in Q2. 10194 tracks in catalog. 50 TRQL Radio USB keys ordered.

May 25th, 2018 : New musical object: the nugget. Algorithm #25 makes usage of the nuggets (48 nuggets developed). All the encoding of the tracks modified: from 32-bit encoding to 16- bit. 11110 tracks in the catalog. Early contacts about a radio for SN Brussels Airlines. 691 members on FB. Start of genre tagging for all tracks. 75% of the tracks are not genre-tagged! New music flows: TRQL Radio is now fully Multi-Stations capable!

July 13th, 2018 : 950 members on FB. New musical object: tapas (music of Ronald) ! 120 nuggets in place. 13808 tracks in catalog. 1412 locomotives. Algorithm #28 for AI to create the playlists. Due to the continuous problems we have gotten into with our streaming solution, we have ported everything to another server: 700h of working without problem/crash. SN Brussels Airlines: no further road. Stop discussions. Start work on the TRQL Beamer (a mult-function device to use in points-of-music such as restaurants, hairdressers, hotel lobbies, etc.). 50% of the catalog remains untagged.

August 18th, 2018 : 15000 tracks in catalog. 1427 locomotives. Looking for a Community Manager. Pierre-François Wuille as CIO of TRQL Radio.

Since then, the catalog of TRQL Radio hasn't stopped growing: 26307 tracks as of today! (computed in real-time)

The stats of the radio, collected from the very beginning, are astounding (and also computed in real-time for accuracy's sake):

  • First date: 22-06-2017
  • Last date: 17-07-2019
  • Total days collected: 757
  • Unique Visitors: 243423
  • Average per day: 6204.79
  • Unique visitors of today (17-07-2019): 7827

… and the radio keeps growing, not only in terms of catalog, visits, FB members (1522 as of 20-12-2018 @15:13), but also in terms of code maturity and flexibility because, @TRQL Radio, everything's code!

Digital Transformations

The radio has been conducted as if it was a Digital Transformation, from day 1.

A Digital Transformation is about 5 pillars:

The 5 pillars of a Digital Transformation
  • Products & Services: our product and our services are fully digital
  • Distribution Channels: web and web only. It is a strong will not to go to the FM wave or to the use of the DAB technology
  • Communication Channels: foremost web.
  • Work Methods and Processes: flat structure with people fully empowered; deliver Business Value asap, even if not perfect; collect feedback then inspect and adapt
  • Shorter Cycles: go as fast as possible; no-regret moves; fail quickly prefer to be the first to be wrong than the last to be right; numerous feedback loops, inside and outside the radio

TRQL Radio is entirely founded on these 5 pillars! We made these pillars our DNA, and we turned it into a somewhat actionable Mission Statement:

TRQL Radio's Mission Statement

All this turns to be extremely Agile! But one must be aware of the dangers of being agile, which is depicted here in the Agile Turmoil Cloud of Words:

The Agile Turmoil

TRQL Radio knows how difficult this path can be for organizations that have little experience in that domain. The case of RTBF is an exemplary case (see also Un projet DOCTIRIS pour étudier la transition de la RTBF vers un mode d’organisation « agile »)

This is also why we are slowly but steadily documenting our methodology. The first page can already be found TRQL Radio's website: we need to turn ± 600 pages of explanations and schemas before we can claim Job Done! Stay tuned!

End-to-End Stream

TRQL Radio is conducted by people with the help of … machines! Many parts of our global process (not all though!) are automated and some use Artificial Intelligence.

236 programs run at their own will to achieve most of the automation work needed: make sure the end-to-end process is carried on in the most robust way.

If you take a look at the end-to-end process, there are 3 stages we get into:

End-to-End Process View

Diving into each odf the stages we distinguish 6 main processes we need to tackle every day:

  1. Get Music
  2. Process Music
  3. Artists' Information Harvest
  4. Playlist Generation
  5. Playout
  6. Feedback Loops

1. Get Music

Here is what we mean by «get music»:

  • Look for new releases
  • Look for new artists
  • Categorize discoveries
  • Selection
  • Acquisition

2. Process Music

Here is what we mean by «processing music»:

  • Categorization
  • Energy level, mood, bpm, …
  • Quarantine
  • Genre
  • Style
  • Cover
  • Leveling and volume adjustment
  • Copyright
  • Keywords
  • Lyrics (get and understand)
  • Language detection
  • Start and end detection / start and end normalization
  • Intro detection
  • Rating

3. Artists' Information Harvest

This part of the chain is devoted to the information about Artists and Releases discovered in phase 1 (not only the ones we have selected)

  • Gather information
  • Compare information
  • Store information
  • Digest information
  • Build information representation
  • Constant Source Quality Appraisal

4. Playlist Generation

The generation of playlists is the trickiest part of TRQL Radio. What is requested from the computer is to select music automatically by taking into account a myriad of elements: is track B compatible with track A? has track X been proposed on air yesterday? what jingles should be inserted? when? is this a special day – Apocalypse day? Zakouskis day? Phoebus Terminus Delta day? etc. etc.

That leads to a number of possibilities that seems beyond our human capability. Selecting tracks so that the program that's proposed on air seems natural, well thought, appropriate, varied, etc. is an ENORMOUS effort.

To quibble in a catalog of 26.307 tracks potentially leads to no less than 346.015.971 relations to compute! And that is for ONLY 1 Radio Station.

Potentially, we need to be able to do this for as many radio stations and music flows we take care of. This is where we use the most of our Artificial Intelligence algorithms.

The selection of the music is not easy task, but one that's no less complex is to sequence the music: give it a specific order.

When the order is given, we now must make sure that the core of our musical format is well respected and that it permeates on air! That's when we insert music objects that are recentering the musical line: the central lane of our musical highway.

At this stage, the playlist is examined in terms of attractiveness and as soon as we feel it drops under a certain level, we insert what we call a loco (a classic) to recenter the interest of the listener.

Then comes the sprinkling of jingles whose subject or theme must be appropriate to the track that they precede.

We then have the check whether the playlist is long enough to fit the necessary length: a full day of music is about 86.400 sec. If for any reason, the playlist is not long enough, music is added.

And last but not least, the playlist is finally analyzed globally to make sure we haven't created unwanted grooves (e.g. long sequences of the same music genre that may be very coherent but that creates boredom): the Playlist Analysis! As a real time example of this analysis, please consider the analysis of the playlist of the day:

558 tracks found (jingles included)

Colorful Representation of the Playlist

Duration

30:30:41

What this analysis reveals is a perfect color segueing: it must match our Camelot Wheel algorithm #24 that categorizes music:

The last touch concerns the generation of a cover for the playlist. It may be considered as ancillary, and it is, but when it comes down to publish the playlist on a website, or on social networks, it becomes very handy!

All this work is usually done between 6 seconds and 6 minutes for a playlist of a full day, depending on a number of factors (the most important factor is the novelty of the catalog: the newer the catalog, the bigger the chances are that a lot of "calculations" performed in the past must be recomputed and the longer it takes to generate a playlist)

Playlist generation is automatic. It can also be triggered manually.

5. Playout

The Playout is a phase during which we simply play what has been prepared before. Dumb play? Not so! There is a part in our Playout where we use specific algorithms to mix tracks together. This is based on a 4 x 4 matrix that categorizes starts and ends. So, to put it short, the mix of Track A depends on Track B! It is not always the same sort of mix.

6. Feedback Loops

Knowing what is delighting listeners requires to have feedback loops. A recurrent visitor is one way to check whether we are doing well or not. Collecting Likes and Dislikes is another way of checking whether the music that we propose is pleasing our audience.

Feedback loops are established in all the points of contacts we have: Playout, Suggestion Box, Search interface, Social network (essentially facebook for the time being), our own website, etc.

The objective analysis of these feedbacks influences directly our algorithms so that the music we propose is slightly adapted to such actionable metrics.


A bit more …

This site was conceived and developed by TRQL Labs!

Website Stats

Links

  1. About us
  2. Announcements
  3. Partners
  4. TRQL Labs
  5. API Explorer

Almost Perfect Adult Contemporary Music

Logo du site 'Quitus'

Top

Notre site utilise des cookies pour sauver vos préférences et RIEN d'autre. Nous vous demandons de bien vouloir accepter les cookies qui sont positionnés sur notre site et nous vous promettons en retour de ne RIEN laisser transparaître de votre parcours de visite, dont d'ailleurs vous nous faites l'honneur. Au cas où néanmoins vous n'accepteriez pas d'utiliser les cookies, vous aurez probablement une navigation LÉGÈREMENT dégradée sur notre site. Bonne visite!X