-7

I am working on a project in my college. What I need to is create navigation system. I am using google maps for navigation. But for input I want something that can convert English speech to english text. Means is there a shield for this task that I can use ? Or anything that I can use ?

omerjerk
  • 127
  • 1
  • 7
  • 1
    As you didn't mention any constraints, you can use IBM Watson, it understands natural spoken language. It can convert to text as well. – Chetan Bhargava Mar 20 '13 at 18:55
  • 1
    @ChetanBhargava: Actually, that's not true. Watson understands *written* natural language, but they didn't build speech recognition into the system, at least not for the Jeopardy demonstration. – Dave Tweed Mar 20 '13 at 19:37
  • @DaveTweed "Watson understands spoken language and can answer ad hoc questions about knowledge in its domain. Based on IBM's DeepQA technology--which RPI and seven universities help IBM develop as an open architecture--Watson can quickly sift through vast amounts of unstructured "Big Data" drilling-down to the fundamental answers to the queries put to it by humans." - [From EETIMES](http://www.eetimes.com/electronics-news/4406135/IBM-s-Watson-goes-to-college) – Chetan Bhargava Mar 20 '13 at 20:15
  • Although this question has been closed (perhaps because it isn't ideal for Electronics.SE), there are multiple *limited-command-set* speech-recognition shields out there, the most commonly known one being [The EasyVR Arduino Shield](http://www.tigal.com/product/2333) which can be used for this purpose - unless full natural language processing is required. – Anindo Ghosh Mar 21 '13 at 12:58

2 Answers2

2

Speech-to-text is a hard problem that requires substantial computing power. You're not going to find it in a shield. Your best chance is using an operating system with built in speech recognition (Windows or Android).

(This question is better asked in the software stackexchange)

pjc50
  • 46,540
  • 4
  • 64
  • 126
  • 3
    I would point out that not even Apple's "Siri" does speech recognition in the local processor -- the audio is shipped out to servers in "the cloud" and text is returned. – Dave Tweed Mar 20 '13 at 19:40
  • EasyVR [Multi-language speech recognition Arduino shield](http://www.tigal.com/product/2333) <-- Voice recognition, multiple languages, in an Arduino shield. – Anindo Ghosh Mar 21 '13 at 12:52
  • I'm impressed, although that's recognition of a limited command set not full speech-to-text. – pjc50 Mar 21 '13 at 12:54
  • @pjc50 Agreed... The requirement stated by the OP isn't clear on whether NLP is required, or command-control alone. – Anindo Ghosh Mar 21 '13 at 12:58
  • @DaveTweed I suggest that Siri isn't the only game in town. My Android mobile (and Swype software with Dragon Dictate) happily understands me rattling on in my Indian-accented English, and transcribes what I say to almost excellent text, independent of my online connectivity. – Anindo Ghosh Mar 21 '13 at 13:00
  • @AnindoGhosh: I didn't say it wasn't possible. The Dragon products have been around for quite a while and are generally regarded as being quite good. But they require desktop-class machines, and cellphones/tablets have only recently reached that level of performance. The OP was asking in the context of an Arduino project, however. – Dave Tweed Mar 21 '13 at 13:50
  • A bit of research shows that the shield is built around http://www.sensoryinc.com/products/RSC-4x_series.html "Sensory RSC 4128" processor designed for the purpose. (As usual, the shield has more processing power than the Arduino) – pjc50 Mar 21 '13 at 13:54
  • @pjc50 Yes, I find that really cool - An echo of what happened when I first bought a graphics accelerator card that was much more processing-capable than the Intel motherboard that hosted it. Even my current gaming PC graphics card is so mind-blowing, it leaves the poor "main microprocessor" cowering in the corner in a foetal position. – Anindo Ghosh Mar 21 '13 at 14:02
  • @DaveTweed Yep - Neither your example nor mine would fit onto an Arduino itself (current versions anyway) :-) I was just pointing out that Siri's way isn't the only way - albeit on the _other_ type of hardware, i.e. cellphones. – Anindo Ghosh Mar 21 '13 at 14:04
1

The Arduino is too slow for speech recognition. You probably should look at a DSP or a fast microprocessor, and even then you're looking at a complicated task.

Renan
  • 5,090
  • 2
  • 27
  • 45