Tag Archives: speech

Amazon Transcribe 增加許多客製化功能

Amazon TranscribeAWS 提供的語音辨識服務,最近發表了不少可以客製化的功能:「Amazon Transcribe enhances custom vocabulary with custom pronunciations and display forms」。

一個是可以增加字彙,包括他的發音 (不過得透過 IPA 標,這邊會需要學一些東西):

You can give Amazon Transcribe more information about how to process speech in your input audio or video file by creating a custom vocabulary. A custom vocabulary is a list of specific words that you want Amazon Transcribe to recognize in your audio input. These are generally domain-specific words and phrases, words that Amazon Transcribe isn't recognizing, or proper nouns.

Now, with the use of characters from the International Phonetic Alphabet (IPA), you can enhance each custom terminology with corresponding custom pronunciations. Alternatively, you can also use the standard orthography of the language to mimic the way that the word or phrase sounds.

另外是定義詞彙的標示方法:

Additionally, you can now designate exactly how a customer terminology should be displayed when it is transcribed (e.g. “Street” as “St.” versus “ST”).

這對於專有名詞的部份應該是很好用?像是人名...

開源的語音助理 Leon

目前的語音助理都會把資訊傳回到中央伺服器,而 leon-ai/leon 則是希望在本地端解決:

Leon is an open-source personal assistant who can live on your server.

He does stuff when you ask him for.

You can talk to him and he can talk to you. You can also text him and he can also text you. If you want to, Leon can communicate with you by being offline to protect your privacy.

作者錄了一段目前版本的功能,可以看到語音輸入與分析運作的還不錯:

看到 1.0.0 beta,最近應該會釋出正式版,接下來就是看可以掛什麼 plugin 進去讓他更好用?

reCAPTCHA 與語音辨識:以子之矛,攻子之盾

GooglereCAPTCHA 大概是目前最常用的反制機器人工具了,但因為 accessibility 的原因 (而且應該是有法令要求),還是需要提供盲人可以存取的方式,也就是以語音判斷是否是機器人。

unCaptcha2 就是用這塊,加上 Google 自家的語音辨識 API (也支援其他家 API) 可以直接打穿:

現有的程式碼已經先被 Google 反制,但目的是展示出這樣的概念。

Google 發表新的 TTS (Text-to-Speech) 技術 Tacotron 2

Tacotron 是 Google 發表的 TTS 技術 (i.e. 輸入文字,請電腦發音),而前一版的 Tacotron 的錄音可以參考「Audio samples from "Tacotron: Towards End-to-End Speech Synthesis"」,論文則是在「Tacotron: Towards End-to-End Speech Synthesis」這邊可以看到。

這一版的則是在 Twitter 上看到有人提到:

這一版叫做 Tacotron 2,錄音可以參考「Audio samples from "Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions"」,論文在「Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions」。

這次在錄音頁面的最下面提供了盲測 (人類與 Tacotron 2 的錄音),基本上已經分不出哪個是真人了...

Mozilla 實做百度發表的 Speech-To-Text 引擎 Deep Speech

Hacker News 上看到 MozillaGitHub 上的 mozilla/DeepSpeech 這個專案,用 TensorFlow 實做了百度的「Deep Speech: Scaling up end-to-end speech recognition」論文:

A TensorFlow implementation of Baidu's DeepSpeech architecture

語音轉文字的方案,Mozilla 開專案實做出來了...

這程式碼需要安裝 Git Large File Storage 才能完整下載包含訓練資料的部份:

Manually install Git Large File Storage, then clone the repository normally:
git clone https://github.com/mozilla/DeepSpeech

而目前已經有的資料來自於 Mozilla 另外一個專案「Common Voice」:

The Common Voice project is Mozilla's initiative to help teach machines how real people speak.

Common Voice 這個專案目前只有英文,網頁上就可以參與 validation 過程...

用程式自動同步字幕與聲音

Hacker News 上看到的專案,readbeyond/aeneas

aeneas is a Python/C library and a set of tools to automagically synchronize audio and text (aka forced alignment).

馬上想到的是... 這根本就是字幕組的福音 XDDD

支援的語言:

Confirmed working on 38 languages: AFR, ARA, BUL, CAT, CYM, CES, DAN, DEU, ELL, ENG, EPO, EST, FAS, FIN, FRA, GLE, GRC, HRV, HUN, ISL, ITA, JPN, LAT, LAV, LIT, NLD, NOR, RON, RUS, POL, POR, SLK, SPA, SRP, SWA, SWE, TUR, UKR

除了 ENG 以外,有 JPN... XD

用 Google 的 Speech Recognition API 破 Google 的 reCAPTCHA

就是「以子之矛,攻子之盾」的概念,用 Speech Recognition APIreCAPTCHA:「ReBreakCaptcha: Breaking Google’s ReCaptcha v2 using.. Google」。

就算 Google 在 reCAPTCHA 的聲音裡面加入 watermark,讓自家的 Speech Recognition API 拒絕分析,還是有其他家的可以用 (像是 Amazon Lex 或是 Bing Speech API),所以這樣做不是什麼好解法。

Amazon Polly 與 Amazon Lex:人機介面中的語音處理

AWS 這次推出的這兩個服務剛好成對:「Amazon Polly – Text to Speech in 47 Voices and 24 Languages」、「Amazon Lex – Build Conversational Voice & Text Interfaces」。

Amazon Polly 負責把文字唸出來變成語音,而 Amazon Lex 則是將語音辨識回文字,不過目前都還不支援中文... 但畢竟讓 user interface 這塊變得更親民了,算是基礎建設中服務,讓 startup 專心在產品本身上。