Vu que je bosse à mi-temps je me suis mis sur eLance pour arrondir mes fins de mois. J'ai trouvé pour l'instant deux clients sérieux qui payent décemment, et qui sont capable d'exprimer ce dont ils ont besoin, mais j'ai aussi droit à un florilège de tarés. Le dernier en date m'a bien fait marrer aussi je ne resiste pas au plaisir de vous faire partager ceci (zappez sur les deux derniers si vous êtes pressés) :

Lui, m'invitant à répondre à une offre appelée 'Project One' :

Project One (Timing based programming)

This project is a paperwork hardware design, which must be implemented in a C program to work with the FLDIGI software which uses PortAudio. Since this is an open source project, it must be done using Ubuntu 14.04 under Linux. In a hardware design everything must happen in real time, so this C program is time based. Samples are 16 bit numbers that come from the microphone and can go to the headset in blocks of 256 numbers at the same time in both directions. Then the next block get connected so no samples are lost.

When a flag called SIU (sample in use) is set the next number from the input block is copied into a location called INPUT. SIU is cleared after the number at the location OUTPUT is copied into the the next location of the output block. When SIU is cleared, the locations at both blocks are changed to the next locations and then the SIU flag is set. But the SIU flag is not set when the last number in the blocks have been used. Instead a request for the input and output blocks flag is set the RSNU (request for sample not used) flag is cleared. When the flag that denotes the next input and output blocks are ready to be used is set, both blocks start with their first locations and the SIU flag is set.

So now let's make a program to test the SIU flag by copying the number at INPUT to OUTPUT using the SIU flag to make the copy. I also need you to make me understand how the software makes and uses the SIU flag.

Moi, dans ma réponse à l'offre :

Your requirements are unclear to me. For example, you talk about hardware design, yet you want a C program (a software). You talk about real-time, yet you want this to work on Ubuntu Linux, which is not a real-time OS. etc.

In order for me to understand what it is that you want, I think it would be best if we talked on shype or google hangout (***@gmail.com). I'll then be able to tell you if I feel confident I can do this roject, and how much it would cost.

Il me relance via la messagerie interne :

***,

I need to know if you want to become part of out team. All members must complete Project One of 26 projects. I need to judge the greed factor of your fixed price bid. Please get back to me about this.

Change communication for the better

Do you want to be a member of a team that changes communication for the better? Let us take voice, change its bandwidth to 125 Hz, send the digital data to another location and play the original voice.

This open source software project needs a team of people who want to do this. I know how to do the hard part, but I don’t have an algorithm or a codec and I don’t do software.

There are four parts to this project which should be done in order.

1. Change the 16 bit samples from a soundcard at 48,000 Hz sample rate to 16 long float numbers at 125 Hz data rate.

2. Change the 16 long float numbers at 125 Hz data rate to one binary digit at 125 Hz data rate.

3. Modify existing open source software to take that one binary digit at 125 Hz data rate to another location and recover the one binary digit at 125 Hz data rate.

4. Change the recovered binary digit at 125 Hz data rate to the original voice.

Project One is the beginning of this. I don’t know how to run your software, but I want to edit what you send me and have you run it. If we get along, I will make you part of the team.

Je lui répond :

Dear Secret S. [ND : C'est son pseudo. Secret S. zlol]

What you write is exceedingly unclear. If you want my help we really need to discuss in real time, because you seem to work under wrong assumptions about many things.

You cannot transmit voice at 125 bit/s. To the best of my knowledge, the lowest bitrate you need to transmit speech is 800 bit/s. And this is using state of the art encoding.

You need to tell me more about your project, so I can give you information about what is theoretically possible, and what is technically wise.

Unless you have a few Fields medalists in your team and some millions dollars, you'll never transmit speech at 125b/s.

Good luck,

***.

Et il termine :

***,

Thank you for your honesty. I can tell you do not want to be part of the team that does imposible things. When we succeed you will regret your decision.

secret

Si vous avez une idée brillante pour rigoler avec lui, je veux bien poster vos messages, sinon je vais juste l'ignorer.

Dernière modification le 17/03/15 à 15:36 par Linschn

Envoie-lui un mp3 de bonne chance à 125 bps.

Ça risque d'être chaud de trouver un encodeur qui accepte d'essayer de descendre à 125 bps, mais c'est une super idée ! :D

[Ignore, j'ai les doigts carrées].

Dernière modification le 17/03/15 à 15:36 par Linschn

Merci pour l'idée :
url : https://soundcloud.com/linschn/goodluck

Dear secret s.

I wish you success in your endeavour. I truly do. Who wouldn't want to be able to store or transmit information at less than 1/100th of the cost silly people like Shannon said was necessary ? I certainly would.

I can't wait until Netflix streams HD content to my parent's house, given that they only have ~300kbps link. Imagine that ! It's more than 2000 times what you need to transmit audio !

Why haven't you come 20 years earlier ? Do you know how long I had to wait for pictures to load on my old 33kbps link ? Had you been there I could have downloaded two or three internets in a matter of hours.

Anyway I digress. Attached is a mp3 file of me wishing you good luck. I encoded it a 8 kbps (that is 64 times the rate you propose to use), because the encoder I used could not use a lower setting. I hope you'll be amazed by the audio quality.

Please keep me posted on how you progress. Failing that, I'll just stay tuned for the next Turing award announcement.

***.

P.S. More seriously, though, if you tell me what you actually are trying to achieve, I can help you find a realistic way to do it.

Dernière modification le 17/03/15 à 16:00 par Linschn

When we succeed you will regret your decision

haha ^^

Hahaha :D

Linschn a écrit
Thank you for your honesty. I can tell you do not want to be part of the team that does imposible things. When we succeed you will regret your decision.

Unfortunatly, working on my Time Machine is really too time consuming for me to join you right now. However, once I succeed, I'll be able to come back here in 2015 and join teh awesome Pr0ject 0nE. I just can't lose.

Il a répondu !

Shannon is correct about voice. So only transmit the speech and timing then recreate the amplitude and pitch at the receiver by sending them only once before the transmission.

Mouais, ça peut être fait e.g. avec du deep learning : avec un autoencodeur (un réseau de neurone qui apprend à recopier sur la sortie ce qu'on lui donne en entrée) dont la couche du milieu ne contient qu'un petit nombre de neurones, y'aurait moyen de ne transmettre que la valeur des neurones du milieu. Mais c'est pas un projet qu'on balance sur eLance, c'est un projet digne d'un vrai labo, avec plusieurs publis à la clef.

Je viens de fair une passe ultra rapide sur la biblio, y'a aussi des méthodes non neuronales, mais les débits sont bien supérieurs à ses 125bps.

Je vais essayer de le récupérer, avec un peu de chance j'arriverai à placer la Time Machine.

J'ai appris un truc grâce à lui :
"A cepstrum(/ˈkɛpstrəmˈˌˈsɛpstrəmˈ/) is the result of taking the Inverse Fourier transform (IFT) of the logarithm of the estimated spectrum of a signal."

Je croyais que c'était spectrum mal écrit.

Je viens de lui répondre :

Now we are getting somewhere !

Did you do a survey of the literature on this ?

What is the state of the art lower bound on bandwidth to be able to reconstruct speech ?

What learning techniques do you want to use to construct the encoder/deencoder ? And, related, what description of speech do you want to use ?

A cursory google scholar search points to "mel frequency cepstral coefficients" and pitch frequencies. Is that what you had in mind ?

Have you thought about deep learning ? An autoencoder with a small intermediate layer may reduce bandwidth usage.

Do you have a training set of speech sequences ?

If you don't have the answer to these questions. I can help you find the answers.

Le nouveau Zuckerberg !

I know what needs to be done and I am looking for someone to do it. You sound like you want to tell me to do something else. You don't want to be part of the team. You don't want to do it my way. The only things left is that you don't have any ability to make code. But you believe I am wrong. Really?

Faut que je file donner un cours d'"internet" dans l'asso de mon quartier. Je lui répondrai en revenant. Il est très bon, ça serait bien de la jouer repentant pour le faire tenir un peu longtemps sans trop le fâcher.

Dernière modification le 17/03/15 à 17:52 par Linschn

Spa gentil, tu vois bien que c'est un enfant...

T'as raison Gingembre. J'me suis excusé. J'espère qu'il m'en voudra pas trop.

I'll do it sideways if you pay me to.

I was just looking out for you. It is very surprising to me that somebody would be looking for help on elance for a project that would, at the very least, warrant a publication in Nature if it succeeds.

If you tell me that you know what you are doing, who am I to doubt you ? I'll be a paid part of your team on a heartbeat. I'm a freelancer, a mercenary. I'll be your tactical weapon on this project.

I'd like to apologize about any hesitation I may have expressed about this project. I was trying to warn you about possible shortcomings of your vision. But you are sure of yourself, which means you have thought about this a lot, and you are not doing things on a whim.

I'll follow you. I'll code what to tell me to code. I'll implement your protocol, we'll do it your way. I'll have the satisfaction of pushing boundaries, and tackle a real problem. Let's do it !

Dernière modification le 17/03/15 à 21:47 par Linschn

.

Dernière modification le 17/03/15 à 21:47 par Linschn

Un peu gros, non ?

(sauf si c'est Sarkozy en face)

You have to prove to me and the rest of the team that you know how to write code and follow directions. Submit a bid of $1 on Project One and I will give you a chance. All members of the team have to completed Project One before being allowed to bid on any of other 25 projects. The total project is real time based. For example Project Three will have 16 FIR audio filters running constantly almost in parralle.

So now you know what it feels like to be insulted. Make me you best offer for Project One. Remember this is a open source project and I will be paying you as an individual not a business.

Je suis tombé sur Grey. He doesn't do romance. J'espère que j'aurai la chance d'avoir le droit de travailler pour lui.

Blague à part, je vais arrêter là avec ce clown, il vient de montrer son vrai visage : il faut bosser gratuitement.

Oui,on se doutait un peu que c'était là où il voulait arriver ;)
Je suis sur que ça marche en plus...

kaplan a écrit)
Je suis sur que ça marche en plus...

J'ai dû me prostituer sur mes deux premier projets, pour obtenir une évaluation. Ensuite ça va beaucoup mieux, mais se faire engager quand on a pas de feedback c'est mission impossible.

Il me relance, en plus :

***,

So the open sourced scared you off. Did you really think you would get rich from what you contributed? My name is secret. Your name will be on the team. Others may want people who have changed the world in the future for other projects. Are you sure you don't want to be on the team?

secret spftware

MY NAME IS SECRET

Le spin-off de ce fameux western.

Haha :)

Mince au début j'ai cru que j'étais sur le topic spam :x
Mais non, pas -assez- de fautes.