{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "Offline_ASR.ipynb",
"provenance": [],
"collapsed_sections": [],
"toc_visible": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"accelerator": "GPU"
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "_wIWPxBVc3_O"
},
"source": [
"# NeMo offline ASR\n",
"\n",
"This notebook demonstrates how to \n",
"\n",
"* transcribe an audio file (offline ASR) with greedy decoder\n",
"* extract timestamps information from the model to split audio into separate words\n",
"* use beam search decoder with N-gram language model re-scoring\n",
"\n",
"You may find more info on how to train and use language models for ASR models here:\n",
"https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/asr_language_modeling.html\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gzcsqceVdtj3"
},
"source": [
"## Installation\n",
"NeMo can be installed via simple pip command. \n",
"\n",
"Optional CTC beam search decoder might require restart of Colab runtime after installation."
]
},
{
"cell_type": "code",
"metadata": {
"id": "I9eIxAyKHREB"
},
"source": [
"BRANCH = 'r1.17.0'\n",
"try:\n",
" # Import NeMo Speech Recognition collection\n",
" import nemo.collections.asr as nemo_asr\n",
"except ModuleNotFoundError:\n",
" !python -m pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[all]\n",
"\n",
"# check if we have optional Plotly for visualization\n",
"try:\n",
" from plotly import graph_objects as go\n",
"except ModuleNotFoundError:\n",
" !pip install plotly\n",
" from plotly import graph_objects as go\n",
"\n",
"# check if we have optional ipywidgets for tqdm/notebook\n",
"try:\n",
" import ipywidgets\n",
"except ModuleNotFoundError:\n",
" !pip install ipywidgets\n",
"\n",
"# check if CTC beam decoders are installed\n",
"try:\n",
" import ctc_decoders\n",
"except ModuleNotFoundError:\n",
" # install beam search decoder\n",
" !apt-get update && apt-get install -y swig\n",
" !git clone https://github.com/NVIDIA/NeMo -b \"$BRANCH\"\n",
" !cd NeMo && bash scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh\n",
" print('Restarting Colab runtime to successfully import built module.')\n",
" print('Please re-run the notebook.')\n",
" import os\n",
" os.kill(os.getpid(), 9)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "-X2OyAxreGfl"
},
"source": [
"import numpy as np\n",
"# Import audio processing library\n",
"import librosa\n",
"# We'll use this to listen to audio\n",
"from IPython.display import Audio, display"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "zodyzdyTVXas"
},
"source": [
"## Instantiate pre-trained NeMo model\n",
"``from_pretrained(...)`` API downloads and initializes model directly from the cloud. \n",
"\n",
"Alternatively, ``restore_from(...)`` allows loading a model from a disk.\n",
"\n",
"To display available pre-trained models from the cloud, please use ``list_available_models()`` method."
]
},
{
"cell_type": "code",
"metadata": {
"id": "f_J9cuU1H6Bn"
},
"source": [
"nemo_asr.models.EncDecCTCModel.list_available_models()"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "x2LMVI9qqtEV"
},
"source": [
"Let's load a base English QuartzNet15x5 model."
]
},
{
"cell_type": "code",
"metadata": {
"id": "ZhWmR7lbvwSm"
},
"source": [
"asr_model = nemo_asr.models.EncDecCTCModel.from_pretrained(model_name='QuartzNet15x5Base-En', strict=False)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "HESTZmIzzCEj"
},
"source": [
"## Get test audio clip"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "QPWn89l-zLXo"
},
"source": [
"Let's download and analyze a test audio signal."
]
},
{
"cell_type": "code",
"metadata": {
"id": "02gDfK7czSVV"
},
"source": [
"# Download audio sample which we'll try\n",
"# This is a sample from LibriSpeech dev clean subset - the model hasn't seen it before\n",
"AUDIO_FILENAME = '1919-142785-0028.wav'\n",
"!wget https://dldata-public.s3.us-east-2.amazonaws.com/1919-142785-0028.wav\n",
"\n",
"# load audio signal with librosa\n",
"signal, sample_rate = librosa.load(AUDIO_FILENAME, sr=None)\n",
"\n",
"# display audio player for the signal\n",
"display(Audio(data=signal, rate=sample_rate))\n",
"\n",
"# plot the signal in time domain\n",
"fig_signal = go.Figure(\n",
" go.Scatter(x=np.arange(signal.shape[0])/sample_rate,\n",
" y=signal, line={'color': 'green'},\n",
" name='Waveform',\n",
" hovertemplate='Time: %{x:.2f} s
Amplitude: %{y:.2f}
'),\n",
" layout={\n",
" 'height': 300,\n",
" 'xaxis': {'title': 'Time, s'},\n",
" 'yaxis': {'title': 'Amplitude'},\n",
" 'title': 'Audio Signal',\n",
" 'margin': dict(l=0, r=0, t=40, b=0, pad=0),\n",
" }\n",
")\n",
"fig_signal.show()\n",
"\n",
"# calculate amplitude spectrum\n",
"time_stride=0.01\n",
"hop_length = int(sample_rate*time_stride)\n",
"n_fft = 512\n",
"# linear scale spectrogram\n",
"s = librosa.stft(y=signal,\n",
" n_fft=n_fft,\n",
" hop_length=hop_length)\n",
"s_db = librosa.power_to_db(np.abs(s)**2, ref=np.max, top_db=100)\n",
"\n",
"# plot the signal in frequency domain\n",
"fig_spectrum = go.Figure(\n",
" go.Heatmap(z=s_db,\n",
" colorscale=[\n",
" [0, 'rgb(30,62,62)'],\n",
" [0.5, 'rgb(30,128,128)'],\n",
" [1, 'rgb(30,255,30)'],\n",
" ],\n",
" colorbar=dict(\n",
" ticksuffix=' dB'\n",
" ),\n",
" dx=time_stride, dy=sample_rate/n_fft/1000,\n",
" name='Spectrogram',\n",
" hovertemplate='Time: %{x:.2f} s
Frequency: %{y:.2f} kHz
Magnitude: %{z:.2f} dB'),\n",
" layout={\n",
" 'height': 300,\n",
" 'xaxis': {'title': 'Time, s'},\n",
" 'yaxis': {'title': 'Frequency, kHz'},\n",
" 'title': 'Spectrogram',\n",
" 'margin': dict(l=0, r=0, t=40, b=0, pad=0),\n",
" }\n",
")\n",
"fig_spectrum.show()"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "jQSj-IhEhrtI"
},
"source": [
"## Offline inference\n",
"If we have an entire audio clip available, then we can do offline inference with a pre-trained model to transcribe it.\n",
"\n",
"The easiest way to do it is to call ASR model's ``transcribe(...)`` method that allows transcribing multiple files in a batch."
]
},
{
"cell_type": "code",
"metadata": {
"id": "s0ERrXIzKpwu"
},
"source": [
"# Convert our audio sample to text\n",
"files = [AUDIO_FILENAME]\n",
"transcript = asr_model.transcribe(paths2audio_files=files)[0]\n",
"print(f'Transcript: \"{transcript}\"')"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "_UOoj-WfQoL_"
},
"source": [
"## Extract timestamps and split words\n",
"``transcribe()`` generates a text applying a CTC greedy decoder to raw probabilities distribution over alphabet's characters from ASR model. We can get those raw probabilities with ``logprobs=True`` argument."
]
},
{
"cell_type": "code",
"metadata": {
"id": "-0Sk0C9-LmAR"
},
"source": [
"# softmax implementation in NumPy\n",
"def softmax(logits):\n",
" e = np.exp(logits - np.max(logits))\n",
" return e / e.sum(axis=-1).reshape([logits.shape[0], 1])\n",
"\n",
"# let's do inference once again but without decoder\n",
"logits = asr_model.transcribe(files, logprobs=True)[0]\n",
"probs = softmax(logits)\n",
"\n",
"# 20ms is duration of a timestep at output of the model\n",
"time_stride = 0.02\n",
"\n",
"# get model's alphabet\n",
"labels = list(asr_model.decoder.vocabulary) + ['blank']\n",
"labels[0] = 'space'\n",
"\n",
"# plot probability distribution over characters for each timestep\n",
"fig_probs = go.Figure(\n",
" go.Heatmap(z=probs.transpose(),\n",
" colorscale=[\n",
" [0, 'rgb(30,62,62)'],\n",
" [1, 'rgb(30,255,30)'],\n",
" ],\n",
" y=labels,\n",
" dx=time_stride,\n",
" name='Probs',\n",
" hovertemplate='Time: %{x:.2f} s
Character: %{y}
Probability: %{z:.2f}'),\n",
" layout={\n",
" 'height': 300,\n",
" 'xaxis': {'title': 'Time, s'},\n",
" 'yaxis': {'title': 'Characters'},\n",
" 'title': 'Character Probabilities',\n",
" 'margin': dict(l=0, r=0, t=40, b=0, pad=0),\n",
" }\n",
")\n",
"fig_probs.show()"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "YiNMZBodIaSP"
},
"source": [
"It is easy to identify timesteps for space character."
]
},
{
"cell_type": "code",
"metadata": {
"id": "32aaW3HEJ89l"
},
"source": [
"# get timestamps for space symbols\n",
"spaces = []\n",
"\n",
"state = ''\n",
"idx_state = 0\n",
"\n",
"if np.argmax(probs[0]) == 0:\n",
" state = 'space'\n",
"\n",
"for idx in range(1, probs.shape[0]):\n",
" current_char_idx = np.argmax(probs[idx])\n",
" if state == 'space' and current_char_idx != 0 and current_char_idx != 28:\n",
" spaces.append([idx_state, idx-1])\n",
" state = ''\n",
" if state == '':\n",
" if current_char_idx == 0:\n",
" state = 'space'\n",
" idx_state = idx\n",
"\n",
"if state == 'space':\n",
" spaces.append([idx_state, len(pred)-1])"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "rqg4oxpsL8cW"
},
"source": [
"Then we can split original audio signal into separate words. It is worth to mention that all timestamps have a delay (or an offset) depending on the model. We need to take it into account for alignment."
]
},
{
"cell_type": "code",
"metadata": {
"id": "a-LSg9dSL_O1"
},
"source": [
"# calibration offset for timestamps: 180 ms\n",
"offset = -0.18\n",
"\n",
"# split the transcript into words\n",
"words = transcript.split()\n",
"\n",
"# cut words\n",
"pos_prev = 0\n",
"for j, spot in enumerate(spaces):\n",
" display(words[j])\n",
" pos_end = offset + (spot[0]+spot[1])/2*time_stride\n",
" display(Audio(signal[int(pos_prev*sample_rate):int(pos_end*sample_rate)],\n",
" rate=sample_rate))\n",
" pos_prev = pos_end\n",
"\n",
"display(words[j+1])\n",
"display(Audio(signal[int(pos_prev*sample_rate):],\n",
" rate=sample_rate))"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "Q8Jvwe4Ahncx"
},
"source": [
"## Offline inference with beam search decoder and N-gram language model re-scoring\n",
"\n",
"It is possible to use an external [KenLM](https://kheafield.com/code/kenlm/)-based N-gram language model to rescore multiple transcription candidates. \n",
"\n",
"Let's download and preprocess LibriSpeech 3-gram language model."
]
},
{
"cell_type": "code",
"metadata": {
"id": "EIh8wTVs5uH7"
},
"source": [
"import gzip\n",
"import os, shutil, wget\n",
"\n",
"lm_gzip_path = '3-gram.pruned.1e-7.arpa.gz'\n",
"if not os.path.exists(lm_gzip_path):\n",
" print('Downloading pruned 3-gram model.')\n",
" lm_url = 'http://www.openslr.org/resources/11/3-gram.pruned.1e-7.arpa.gz'\n",
" lm_gzip_path = wget.download(lm_url)\n",
" print('Downloaded the 3-gram language model.')\n",
"else:\n",
" print('Pruned .arpa.gz already exists.')\n",
"\n",
"uppercase_lm_path = '3-gram.pruned.1e-7.arpa'\n",
"if not os.path.exists(uppercase_lm_path):\n",
" with gzip.open(lm_gzip_path, 'rb') as f_zipped:\n",
" with open(uppercase_lm_path, 'wb') as f_unzipped:\n",
" shutil.copyfileobj(f_zipped, f_unzipped)\n",
" print('Unzipped the 3-gram language model.')\n",
"else:\n",
" print('Unzipped .arpa already exists.')\n",
"\n",
"lm_path = 'lowercase_3-gram.pruned.1e-7.arpa'\n",
"if not os.path.exists(lm_path):\n",
" with open(uppercase_lm_path, 'r') as f_upper:\n",
" with open(lm_path, 'w') as f_lower:\n",
" for line in f_upper:\n",
" f_lower.write(line.lower())\n",
"print('Converted language model file to lowercase.')"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "fLDbUkzzUAqW"
},
"source": [
"Let's instantiate ``BeamSearchDecoderWithLM`` module."
]
},
{
"cell_type": "code",
"metadata": {
"id": "_qgKa9L954bJ"
},
"source": [
"beam_search_lm = nemo_asr.modules.BeamSearchDecoderWithLM(\n",
" vocab=list(asr_model.decoder.vocabulary),\n",
" beam_width=16,\n",
" alpha=2, beta=1.5,\n",
" lm_path=lm_path,\n",
" num_cpus=max(os.cpu_count(), 1),\n",
" input_tensor=False)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "NSH8EvL7USac"
},
"source": [
"Now we can check all transcription candidates along with their scores."
]
},
{
"cell_type": "code",
"metadata": {
"id": "nV1CAy0Dit-g"
},
"source": [
"beam_search_lm.forward(log_probs = np.expand_dims(probs, axis=0), log_probs_length=None)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"# Greedy Decoding Time Stamps\n",
"\n",
"While the above approach works well for character based CTC models, it requires careful tuning of offset parameter as well as computation of the word time stamp offsets.\n",
"\n",
"We therefore provide a simple way to obtain greedy decoding word time stamps directly using the familiar \"model.transcribe()\" method, which works quite well for character and subword models.\n",
"\n",
"**Note**: We find that larger models that have converged to strong scores on the dataset usually have better word alignments. If evaluated on a completely out of domain audio sample, it might produce very poor time stamps."
],
"metadata": {
"id": "LPtMzLE4T7T-"
}
},
{
"cell_type": "code",
"source": [
"from omegaconf import OmegaConf, open_dict"
],
"metadata": {
"id": "z_0pO-TaUIHU"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"For the purposes of this demonstration, we will use Conformer CTC Large, a 120 M parameter model trained on thousands of hours of English speech."
],
"metadata": {
"id": "i0Epb8D-rW3-"
}
},
{
"cell_type": "code",
"source": [
"asr_model_subword = nemo_asr.models.ASRModel.from_pretrained(\"stt_en_conformer_ctc_large\")"
],
"metadata": {
"id": "Ky7OpuikbBTb"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"## CTC Decoding Strategy\n",
"\n",
"NeMo CTC models have an internal decoding strategy that can be updated after training. In our case, we will enable the greedy decoding step to compute word time stamps, as well as preserve the log probability predictions."
],
"metadata": {
"id": "vwN6wddTrhno"
}
},
{
"cell_type": "code",
"metadata": {
"id": "ubpcxp6z3ZF-"
},
"source": [
"decoding_cfg = asr_model_subword.cfg.decoding\n",
"print(OmegaConf.to_yaml(decoding_cfg))"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"decoding_cfg.preserve_alignments = True\n",
"decoding_cfg.compute_timestamps = True\n",
"asr_model_subword.change_decoding_strategy(decoding_cfg)"
],
"metadata": {
"id": "pKUsMlUbUAxv"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"Next, we simply transcribe the audio file, and pass the flag `return_hypotheses=True`. This will return a list of `Hypothesis` objects instead of the predicted text."
],
"metadata": {
"id": "EdX0Drncr8Yl"
}
},
{
"cell_type": "code",
"source": [
"hypothesis = asr_model_subword.transcribe([AUDIO_FILENAME], return_hypotheses=True)[0]"
],
"metadata": {
"id": "SUkfIyYzUbaB"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"print(\"Greedy prediction :\", hypothesis.text)"
],
"metadata": {
"id": "duaxOSPXUmQ0"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"## Hypothesis - Time Stamps\n",
"\n",
"Since we previously set the flag for `decoding_cfg.compute_timestamps`, the hypothesis now contains a dictionary in it, accessed via `hypothesis.timestep`. This dictionary contains multiple useful lists, detailing the time step at which some token was predicted, the character / subword / word time stamps."
],
"metadata": {
"id": "_5hfsiDGsM19"
}
},
{
"cell_type": "code",
"source": [
"timestamp_dict = hypothesis.timestep\n",
"print(\"Hypothesis contains following timestep information :\", list(timestamp_dict.keys()))"
],
"metadata": {
"id": "vh7K_9D1UrQp"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"# 40ms is duration of a timestep at output of the Conformer\n",
"time_stride = 4 * asr_model_subword.cfg.preprocessor.window_stride\n",
"\n",
"##################################################################\n",
"\n",
"word_timestamps = timestamp_dict['word']\n",
"\n",
"for stamp in word_timestamps:\n",
" start = stamp['start_offset'] * time_stride\n",
" end = stamp['end_offset'] * time_stride\n",
" word = stamp['char'] if 'char' in stamp else stamp['word']\n",
"\n",
" print(f\"Time : {start:0.2f} - {end:0.2f} - {word}\")\n",
" display(Audio(signal[int(start * sample_rate) : int(end * sample_rate)], rate=sample_rate))"
],
"metadata": {
"id": "fogttpCTVTEZ"
},
"execution_count": null,
"outputs": []
}
]
}